Sakana AI, an artificial intelligence startup based in Tokyo, made waves with its release of AI models on Wednesday. Founded by two ex-Google researchers, the company introduced a novel approach to model development, drawing inspiration from evolutionary principles akin to breeding and natural selection.
The technique, termed “model merging,” involves combining existing AI models to create new ones. This process, infused with evolutionary methods, generates hundreds of model generations.
Successive iterations identify the most effective models, which then serve as the “parents” for the next generation.
According to Sakana AI founder David Ha, the company is unveiling three Japanese language models, with two models being open-sourced. Ha, along with fellow co-founder Llion Jones, brings a wealth of experience from their tenure at Google.
Jones notably contributed to Google’s seminal 2017 research paper “Attention Is All You Need,” which introduced the transformative “transformer” deep learning architecture.
This innovation laid the groundwork for the widely acclaimed chatbot ChatGPT, sparking a race to harness generative AI for various applications. Ha, formerly head of research at Stability AI and a Google Brain researcher, adds valuable expertise to the team.
The departure of all authors of the groundbreaking Google paper signaled a shift in the landscape, with former researchers embarking on new ventures.
Venture capitalists have shown keen interest, injecting substantial funding into startups like Character.AI and Cohere, helmed by notable figures such as Noam Shazeer and Aidan Gomez, respectively.
Sakana AI’s vision extends beyond innovative AI development. The company aims to position Tokyo as a prominent AI hub, following in the footsteps of San Francisco’s OpenAI and London’s DeepMind.
In January, Sakana AI announced securing $30 million in seed financing, led by Lux Capital, marking a significant milestone in its journey toward AI excellence.
