Google Brain Researchers Launch AI Startup in Japan

A former Google Brain researcher, Llion Jones, renowned for his co-authorship of the groundbreaking paper – Attention is All You Need has united forces with ex-colleague David Ha to establish Sakana AI, an AI startup headquartered in Tokyo.

Jones, hailing from Wales, recently concluded his tenure at the renowned US tech behemoth, and alongside Ha, the former head of Google’s AI research division in Japan, he has embarked on this pioneering venture as the CTO. Ha, now serving as the CEO of Sakana AI, most recently orchestrated innovative research endeavours at Stability AI, a prominent player in image AI technology.

Sakana AI’s pivotal undertaking revolves around the development of an indigenous generative AI model, characterised by its capability to produce diverse forms of content such as text, images, code, and multimedia elements. This strategic move places Sakana AI in direct competition with titans of the AI realm, including industry giants like Google, Microsoft, and OpenAI, as well as burgeoning startups like Cohere, Character.ai, and Anthropic. 

The name “Sakana,” derived from the Japanese term さかな (sa-ka-na) signifying “fish,” is emblematic of the co-founders’ vision. They envision their enterprise as a collective entity, akin to a school of fish coalescing and functioning in harmony through elementary rules, as explained by Ha on X.

Jones and Ha’s inspiration stems from natural concepts like evolution and collective intelligence, a motif that they seek to embody in their research.

Notably, Jones was a pivotal contributor among a group of eight Google researchers responsible for crafting the Transformers in 2017, which is applicable in almost every major generative AI development such as ChatGPT and Bard for textual generation, and Stability AI, Midjourney, and Dall-E for image synthesis. All of the researchers of the paper have now left Google.

In Jones and Ha’s assessment, existing AI models exhibit constraints by virtue of their rigid, unyielding structures resembling buildings or bridges. This stands in stark contrast to natural systems, which showcase adaptability and sensitivity to external shifts. The co-founders aspire to harness the principles of evolutionary computing to develop AI models that embody these adaptive traits, concurrently addressing concerns pertaining to cost and security.

Having been established in Japan for an extended period, the co-founders strategically chose Tokyo as the epicentre of their operations. Interestingly, Japan has been quite liberal when it comes to deciding the copyright laws in the country.

Furthermore, Jones and Ha anticipate Tokyo’s role as a pivotal hub for training data and tailored model infrastructure to cater to non-western cultures and societies. They perceive this confluence as a catalyst that will propel the next wave of technological breakthroughs. 

Interestingly, OpenAI is also planning to expand to Japan soon.