Llion Jones and David Ha launch their own AI start-up, Sakana AI
Two former Google AI researchers, whose work produced AI bots such as OpenAI’s ChatGPT, Google Bard, Stability AI, Midjourney, and Dall-E, have teamed up to launch their artificial intelligence start-up in Tokyo.
Welshman Llion Jones, originally from Wales, left Google this month and joined hands with his former colleague from Google, David Ha. Previously David was the head of Google’s AI research arm in Japan. He is now the Chief Executive Officer of their Tokyo-based start-up, Sakana AI (an R&D company on a quest to create a new foundation model based on nature-inspired intelligence). He was also involved in the former Stability AI project, an image AI company in London, UK.
How Sakana AI Started
Llion Jones, Sakana’s chief technology officer, was also one of the brains behind building the software named Transformer, which played a major role in developing generative AI. These included chatbots like ChatGPT and Bard and image generators like Stability AI, Midjourney and Dall-E.
Transformer’s first research paper was published in the year 2017. Since then, all of its co-authors have left Google to pursue their own ventures as the worldwide competition for creative AI talent intensifies. Moreover, Jones was the last person to exit Google.
Sakana AI will build its own generative AI model that will be able to produce text, images, code and other multimedia. It will also compete against the world’s primary AI companies like Google, Microsoft, OpenAI, and start-ups such as Cohere, Character.ai and Anthropic, especially in a highly competitive and exceptionally well-funded field.
Funding Required for Sakana AI
In 2023, Microsoft has been able to invest at least $10 bn in its Open AI project, whereas Cohere and Character.ai have been able to fund $2bn and $1bn, respectively, in the last few months.
According to the two co-founders, whose research is motivated by ideas from nature, such as evolution and collective intelligence, the name Sakana, derived from a Japanese word for fish (sa-ka-na), seeks to evoke the idea of “a school of fish coming together and forming a coherent entity from simple rules”.
Current AI Models Have Limits
Jones and Ha assert that the design of present AI models as brittle, immutable structures, such as bridges or buildings, is the source of their limits. Compared to natural systems, when collective intelligence is involved, “are much more sensitive and responsive to environmental changes. Natural systems change and integrate with their surroundings”, they commented. These evolutionary computing ideas will be used to create AI models to solve challenges like system cost and security.
Because it is so difficult to find researchers who can create generative AI software, particularly in North America, the founders have been stationed in Japan for many years. They said they chose the country’s capital, Tokyo, as the company’s headquarters partly because of this.
Tokyo is ideally located for developing an AI business because of its top-notch technical infrastructure and skilled workforce. It is also a multicultural metropolis that attracts top international talent. Moreover, they are also excited to utilise the training data and models tailored for non-western cultures to drive the next technological breakthrough.