Sakana AI's Evolutionary Approach: Tokyo Startup Disrupts AI Landscape

Tokyo-based artificial intelligence startup, Sakana AI, established by former Google researchers, introduced groundbreaking AI models on Wednesday (March 20th).

These models, developed through an innovative method inspired by evolution, leverage a process akin to breeding and natural selection.

Employing a technique termed "model merging," Sakana AI combines existing AI models to generate new ones.

This process, infused with evolutionary principles, results in the creation of numerous model generations.

Successful models from each iteration are identified as the "parents" of subsequent generations, ensuring continuous refinement and advancement.

Founder David Ha revealed that the company is rolling out three Japanese language models, with two of them being open-sourced.

Ha, alongside co-founder Llion Jones, both renowned former Google researchers, brings extensive expertise to Sakana AI.

Jones notably contributed to Google's 2017 research paper "Attention Is All You Need," which introduced the transformative "transformer" deep learning architecture.

This innovation laid the groundwork for widely acclaimed products like ChatGPT, igniting a race to develop generative AI-powered solutions.

Ha, formerly heading research at Stability AI and serving as a Google Brain researcher, underscores a trend where all authors of the pioneering Google paper have departed the organization.

Sakana AI's pioneering approach marks a significant leap forward in the realm of artificial intelligence, promising innovative solutions and advancements in the field.

Source: Anna Tong / Reuters