Sakana AI has announced that researcher Jürgen Schmidhuber will join the company as Chief Scientific Advisor. He will lead the RSI Lab, an organization dedicated to researching "Recursive Self-Improvement (RSI)" where AI improves itself.
NewsSakana AIJürgen Schmidhuber
Jürgen Schmidhuber Joins Sakana AI to Lead RSI Research Organization
This article is a translation. Read the Japanese original
Schmidhuber is a well-known researcher, recognized as a co-developer of "LSTM," a neural network used for processing sequential data such as text and audio. Sakana AI CEO David Ha stated that Schmidhuber and others are credited with presenting the prototypes of linear Transformer in 1991, as well as unsupervised pre-training, distillation, and residual learning.
RSI Lab aims to enhance machine intelligence through a recursive cycle where scientific discoveries lead to further discoveries. It is noted that his 1987 doctoral thesis influenced the development of technologies such as the "Darwin Gödel Machine," an AI agent capable of rewriting its own code.
Moving forward, Sakana AI aims to advance the development of "World Models" that simulate the impact of actions on the physical world, targeting the development of Physical AI for manufacturing and robotics. Schmidhuber commented on Japan's strengths in both fundamental neural network technology and advanced robotics, expressing his desire to serve as a bridge between the two.
Sources
- 「現代AIの父」シュミットフーバー氏がSakana AIに参画 RSI研究のアドバイザーに （ITmedia AI+、2026-09-25）
- 公式ブログ（Sakana AI）