Former Google researchers who wrote a paper that sparked the generative AI revolution have teamed up with an ex-colleague in Tokyo to launch a start-up for artificial intelligence.
Sakana AI was founded by David Ha, a former Google AI researcher in Japan, and Welshman Llion Jones who quit the US tech giant earlier this month. Ha, the CEO, was most recently the head of research at Stability AI, an image AI company.
Jones, Sakana’s chief technology officer was one of the eight Google researchers that collaborated to build a software called the transformer. This software underpinned the rise in generative AI including chatbots like ChatGPT, Bard and image generators like Stability AI. Midjourney, and Dall-E.
The research paper on Transformers was published for the first time in June 2017. All of the co-authors left Google since then, mostly to start their own companies as the global race for talent in generative AI heats up. Jones was the eighth and final Google employee to leave.
Sakana AI is building its own model of generative AI — software which can generate text and images, as well as code, code, and other multimedia. Sakana AI will compete with some of the biggest AI companies in the world, including Google, Microsoft and OpenAI. It will also face start-ups like Cohere and Character.ai, as well as other highly funded and competitive AI companies. Microsoft invested $10bn this year in OpenAI in a multi-year deal. Cohere, Character.ai, and Anthropic have raised money in the last few months with valuations of $1bn and $2bn.
According to the co-founders of Sakana, the name is derived from the Japanese word sakana for fish. It evokes the idea of a “school of fish” coming together to form a coherent entity based on simple rules.
Jones and Ha claim that current AI models are limited by the fact that they have been engineered to be brittle structures that cannot be changed, such as bridges or buildings.
Comparatively, natural systems where collective intelligence is a factor, are “very sensitive and responsive” to the changes that occur in their environment. “Natural systems adapt to their environment and become a part of it,” they stated. The hope is to develop AI models based on evolutionary computing principles, which will address cost and security issues.
They said that the founders have been based in Japan since several years and chose its capital to be the headquarters of the company partly because the market is competitive for researchers who are able to build generative AI, especially in North America.
Tokyo’s high-quality infrastructure and highly educated workforce make it the perfect place to grow an AI company. Tokyo is a global metropolis that attracts foreign talent.
They added that “training data and machinery around models tailored to perform in non-western cultures and societies will be a catalyst for the next technological breakthrough.”