The newly formed Recursive Self-Improvement Lab in Tokyo aims to trigger a compounding cycle of scientific discovery to improve machine intelligence. Jürgen Schmidhuber will lead this effort as Sakana AI's Chief Scientific Advisor, directing the development of agent-native world models and recursive self-improvement for physical AI.

Schmidhuber has spent nearly 40 years researching how machines learn to learn, pioneering meta-learning and world models in the 1990s when compute was 1 million times more expensive than current standards. His early work in deep learning inspired existing projects at the firm, including the Darwin Gödel Machine and The AI Scientist, as Sakana AI seeks to recruit more experts for its research hub in Japan.