Sakana AI hires Schmidhuber to steer new self‑improvement lab
The Tokyo company said Jurgen Schmidhuber joins as chief scientific advisor and will guide a lab working on systems that rewrite their own code.
Sakana AI said on Thursday that Jurgen Schmidhuber has joined as chief scientific advisor. He will guide a new lab the Tokyo company calls the Recursive Self-Improvement Lab, and keeps his existing positions.
The lab's stated aim is AI that can rewrite its own code while acting in the physical world. Sakana AI put that description in its announcement. It has not said how large the lab is, or when it expects to publish anything.
Sakana AI credited Schmidhuber with work it says its projects rest on. It listed world models in 1990, deep learning techniques in 1991, and a 1987 thesis on meta-learning and recursive self-improvement. It named the Darwin Godel Machine and The AI Scientist as projects his ideas shaped.
"Japan is a birthplace of foundational neural network architectures and advanced robotics," Schmidhuber said in the announcement. "The future of intelligence is not just language; it is physical AI powered by World Models." The company said it wants to draw researchers to Tokyo.
Sakana AI said it plans to build what it calls Agent-Native World Models for physical AI, and pointed at Japan's manufacturing and robotics industries. It described simulations that could predict physical consequences before a machine acts.