Tokyo’s Sakana AI, co-founded by ex-Google researchers David Ha and Llion Jones, has unveiled AI models developed through a pioneering technique inspired by evolutionary processes. This method, known as “model merging,” involves the fusion of existing AI models to generate new iterations over numerous generations, mimicking natural selection by choosing the most successful models to lead subsequent generations.
The startup, aiming to establish Tokyo as a major AI hub akin to OpenAI’s impact on San Francisco, has open-sourced two of its three Japanese language models. Ha, with a background as Stability AI’s head of research and a Google Brain researcher, along with Jones, known for co-authoring the influential 2017 Google paper “Attention Is All You Need,” are steering Sakana AI’s direction. This paper laid the groundwork for the transformer deep learning architecture, essential for the development of generative AI products like ChatGPT.
With all authors of the transformative Google research now pursuing new ventures, Sakana AI has secured $30 million in seed funding from Lux Capital, highlighting the tech community’s investment in the next wave of AI innovations.
