Sakana AI, Founded by Ex-Google Researchers, Unveils Open-Source Models in Tokyo

Sakana AI, a new artificial intelligence research company headquartered in Tokyo, Japan, created and launched AI models on Wednesday, March 20. This startup AI firm was founded by two former distinguished researchers of Google Inc.

The company said the AI models were built through a new process inspired by evolution. This method was described as similar to breeding and natural selection, as per Reuters.

How the AI Models Were Formed

Sakana AI revealed that a technique called "model merging" was applied to create the new AI models. As the name suggests, this procedure incorporates the existing AI models to generate a new one and then compound this with an approach inspired by evolution.

The entire procedure will produce hundreds of model generations. Sakana AI said that the most successful designs from each generation were identified, and these will be labeled as the "parents" of the next generation of AI.

Researchers Behind the AI Models

Sakana AI is set to launch three Japanese-language models, two of which are open-sourced. The company's founders are David Ha and Llion Jones, both former Google researchers.

The latter authored the "Attention Is All You Need" research paper, published under the tech giant's name in 2017. In addition to working at Google, he previously led Stability AI's research team.

Meanwhile, Nikkei Asia reported the three generative AI models: a large language model (LLM), an image generation model, and an image-to-text artificial intelligence model that uses the "nature-inspired" mechanism to get results. These are said to be the first of their kind in the world.

Sakana AI is now the first AI company to invent this novel way of building generative AI models that use evolution. The company says this method is a cheaper and faster alternative for AI development.

Photo by: Sakana AI Website