Japan’s Sakana AI launches open-source AI models inspired by evolution

AI-inspired by natural selection: Sakana AI, a Tokyo-based Artificial Intelligence (AI) startup founded by two former Google researchers, announced on Wednesday that the release of AI models developed using an innovative method inspired by evolution, akin to breeding and natural selection.

Employing a technique called "model merging," Sakana AI combines existing AI models to generate a new model, incorporating an evolutionary-inspired approach that results in the creation of multiple model generations. The most successful models from each generation are then selected to serve as the "parents" for the subsequent generation.

According to Sakana AI founder David Ha, the company is introducing three Japanese language models, with two of them being open-sourced. The founders, Ha and Llion Jones, both previously worked at Google. Jones co-authored Google's influential 2017 research paper "Attention Is All You Need," which introduced the "transformer" deep learning architecture, laying the groundwork for ChatGPT, a widely-used chatbot. Ha served as the head of research at Stability AI and was also a researcher at Google Brain.

Following their departure from Google, the founders have attracted major investment for their new ventures. Notable investments include AI chatbot startup Character.AI, led by Noam Shazeer, and large language model startup Cohere, founded by Aidan Gomez.

Sakana AI aims to establish Tokyo as a prominent AI hub, similar to how OpenAI and DeepMind have done for San Francisco and London, respectively. In January, Sakana AI announced securing $30 million in seed financing, led by Lux Capital.

(With Reuters inputs)