Founded by former Google researchers, Sakana AI has released open-source AI models

Using a unique evolutionary-inspired approach, Sakana AI, an AI startup in Tokyo founded by former Google researchers, has introduced open-source AI models.

By combining pre-existing AI models with an evolutionary approach, Sakana AI's "model merging" technique creates hundreds of model generations.

After that, the most effective role models from each generation were chosen to serve as the "parents" of the following one.

David Ha, the founder of Sakana AI, told Reuters online from Tokyo that the company is making the three Japanese language models available, with two of them being open-sourced.

The founders of the company are Llion Jones and Ha, two former Google researchers.

One of Jones's contributions to the 2017 Google research paper "Attention Is All You Need" was the introduction of the "transformer" deep learning architecture, which served as the foundation for the widely used chatbot ChatGPT and sparked a rush to create generative AI-powered products.

Ha was a Google Brain researcher before serving as the head of research at Stability AI.

The groundbreaking Google paper's writers have all since departed the company.

Millions of dollars have been invested by venture capitalists in their new businesses, such as the AI chatbot startup Character.AI led by Noam Shazeer and Aidan Gomez's massive language model startup Cohere.

In the same way that OpenAI did for San Francisco and the business DeepMind did for London before, Sakana AI aims to establish the capital of Japan as a hub for AI. Sakana AI announced in January that Lux Capital has led a $30 million seed financing round for the company.