Sakana AI Creates New Artificial Intelligence Model Inspired By Evolution

Illustration of Sakana AI. (photo: x @sakana_ai)

JAKARTA - Sakana AI, Tokyo-based artificial intelligence (AI) startup, launched a new AI model on Wednesday, March 20. The model is claimed to be built using innovative methods inspired by evolution, similar to crossbreeding and natural selection.

Sakana AI uses a technique called "merging model" that incorporates existing AI models to generate new models. They then combine it with an evolution-inspired approach, leading to the creation of hundreds of model generations.

The most successful model of each generation is then identified, and becomes the "induct" of the next generation.

"We released three Japanese models and two of them open-source," said Sakana AI founder David Ha, quoted by VOI from Reuters.

The company's founders are former Google researchers Ha and Llion Jones.

Jones is one of the authors of a 2017 Google research paper entitled "Attention Is All You Need", which introduces a deep learning "transformer" architecture that forms the basis for ChatGPT viral chatbots, and triggers a race to develop products powered by a genrative AI.

Ha was previously the head of research at AI Stability and Google Brain researchers. All the authors of this innovative Google paper have now left the organization.

The venture investors have poured millions of dollars into funding their new start-ups, such as the startup of the AI chatbot character run by Noam Shazeer, and the Cohere big language model startup founded by forced Gomez.

Sakana AI seeks to make Japan's capital AI center, as OpenAI did for San Francisco and DeepMind's company for London earlier. In January, Sakana AI announced it had raised initial funds of 30 million US dollars led by Lux Capital.