SAN JOSE, March 20 (Reuters) - Sakana AI, a Tokyo-based

artificial intelligence startup founded by two prominent former

Google researchers, released AI models on Wednesday it

said were built using a novel method inspired by evolution, akin

to breeding and natural selection.

Sakana AI employed a technique called "model merging" which

combines existing AI models to yield a new model, combining it

with an approach inspired by evolution, leading to the creation

of hundreds of model generations.

The most successful models from each generation were then

identified, becoming the "parents" of the next generation.

The company is releasing the three Japanese language models

and two are being open-sourced, Sakana AI founder David Ha told

Reuters in online remarks from Tokyo.

The company's founders are former Google researchers Ha and

Llion Jones.

Jones is an author on Google's 2017 research paper

"Attention Is All You Need", which introduced the "transformer"

deep learning architecture that formed the basis for viral

chatbot ChatGPT, leading to the race to develop products powered

by generative AI.

Ha was previously the head of research at Stability AI and a

Google Brain researcher.

All the authors of the ground-breaking Google paper have

since left the organisation.

Venture investors have poured millions of dollars in funding

into their new ventures, such as AI chatbot startup Character.AI

run by Noam Shazeer, and the large language model startup Cohere

founded by Aidan Gomez.

Sakana AI seeks to put the Japanese capital on the map as an AI

hub, just as OpenAI did for San Francisco and the company

DeepMind did for London earlier. In January Sakana AI said it

had raised $30 million in seed financing led by Lux Capital.