Tokyo Startup’s Distributed AI
Two ex-leading researchers from Google’s AI division have introduced Sakana AI, a Tokyo-based startup that adopts a “swarm” methodology by amalgamating multiple compact AI models. Notable highlights include:
- Sakana AI will delve into orchestrating specialized smaller models, drawing inspiration from natural swarms, to achieve capabilities akin to those of massive models.
- The underlying objective of this approach is to curtail data requirements and expenses in contrast to the conventional path of enlarging singular immense models.
- A co-founder, Llion Jones, renowned for contributing to the pioneering transformer architecture that underpins ChatGPT, stands as one of the driving forces behind the startup.
- Of particular interest is how, following the transformative impact of transformers in the realm of AI, attention is now fixed on the innovative potentials these former Google luminaries might unveil. The distributed approach adopted by their startup could potentially ignite fresh avenues for advancing scalability in the ongoing AI competition.