The Pulse
Sakana AI Names Jürgen Schmidhuber Chief Scientific Advisor
Sakana AI announced on September 24 that Jürgen Schmidhuber will join as chief scientific advisor and help guide its Recursive Self-Improvement Lab. The company says the lab is developing its first world models and aims to build systems tha
AI.info Team ·
Schmidhuber joins Sakana AI’s Tokyo research effort
On September 24, Sakana AI announced that Jürgen Schmidhuber will join the company as chief scientific advisor, alongside his current positions. The company says he will help guide its Recursive Self-Improvement (RSI) Lab, a research group focused on using AI to improve the process of developing AI. The announcement places his role within a broader effort to build systems that extend beyond text and act in the physical world.
Schmidhuber described the appointment in terms of that shift. “The future of intelligence is not just language; it is physical AI powered by World Models,” he said. He added that he was joining Sakana AI to help connect Japan’s work on neural network architectures with its advanced robotics.
World models are the lab’s next target
Sakana AI says it is developing its first world models, which the company describes as systems that can simulate the consequences of actions before they happen. The announcement links those models to industrial uses, including planning supply chains and deploying robots without relying as heavily on costly real-world trial and error. Those applications are goals in the company’s post, not capabilities it says are already in operation.
The company says Schmidhuber will travel to Tokyo regularly to work with its team and support the wider Japanese AI ecosystem. It describes his research on world models, meta-learning and the Gödel Machine as relevant to the lab’s direction. The appointment is advisory; Sakana AI’s announcement does not say that he is leaving his existing roles or joining the company full time.
RSI aims to make research feed on itself
Recursive self-improvement refers here to a cycle in which AI systems help research and improve the systems that power them. Sakana AI says the RSI Lab aims to create a compounding cycle of scientific discovery that improves machine intelligence. It describes future systems that could conduct research, make discoveries, rewrite their own code and interact with the physical world.
The company connects that plan to Schmidhuber’s 1987 thesis on recursive self-improvement and meta-learning. Sakana AI says those ideas have influenced projects including the Darwin Gödel Machine and The AI Scientist. The announcement does not provide a development timetable, technical specifications for the planned world models or details about how the lab will evaluate systems that modify their own code.
A research appointment, not a results announcement
The announcement sets out a research agenda rather than a new model release. Sakana AI frames the work around “Agent-Native World Models” and says they should simulate physical consequences before an agent acts, while the RSI Lab would connect such work to automated research and system improvement. The post names no benchmark results or deployed industrial systems tied to this plan.
Schmidhuber’s stated role is to help guide the newly formed lab, and the company says it is recruiting researchers and engineers in Tokyo. The immediate concrete steps are the advisory appointment, regular visits to the team and development of Sakana AI’s first world models.