Schmidhuber Joins Sakana AI as Chief Scientific Advisor as RSI Lab Launches

On Sept 24, Sakana AI officially announced that deep learning pioneer Jürgen Schmidhuber, widely called one of the "fathers of modern AI," has joined the company as Chief Scientific Advisor. The role is part-time, run in parallel with his positions at IDSIA, and he will lead the company's new Recursive Self-Improvement (RSI) Lab. The move was also covered by Nikkei and other media.

Confirmed

• Sakana AI announced Schmidhuber as part-time Chief Scientific Advisor, leading the new RSI Lab focused on recursive self-improvement: AI improving itself in a loop of "scientific discoveries enabling further scientific discoveries."
• The company's post highlighted his decades of work: his 1987 thesis on "machines that learn to learn," and pioneering 1990s research on meta-learning, recursive self-improvement and world models when compute was a million times more expensive. He is also an LSTM co-developer.
• Sakana AI also cited its recent groundwork — LLM² (AI automatically inventing optimization algorithms) and Darwin Gödel Machine (agents rewriting their own code to improve performance) — as directly extending Schmidhuber's research.
• On Sept 26, Sakana AI officially launched the RSI Lab, aiming to rebuild the AI R&D process itself with AI, turning static human-led research into an autonomous self-improving engine; the company noted Japan's compute constraints as motivation.

Why it matters

• Schmidhuber is a co-developer of LSTM and a foundational figure in deep learning; his ideas on meta-learning and recursive self-improvement are key theoretical roots of current autonomous AI research.
• The hire and the RSI Lab's launch show Sakana AI formally betting on recursive self-improvement as its core strategy, moving from human-researcher-dependent development toward autonomous scientific discovery.

2026-09-24 ~ 2026-09-26 · 13 related posts

• [source] Sakana AI names deep learning pioneer Jürgen Schmidhuber as Chief Scientific Advisor — SakanaAILabs · 2026-09-24
• [source] Sakana AI formally launches Recursive Self-Improvement Lab to automate R&D — SakanaAILabs · 2026-09-26

11 near-duplicate retellings: RobertTLange · hardmaru · hardmaru · SakanaAILabs · hardmaru · tkasasagi · AdrienLE · hardmaru · hardmaru · wavefnx · AdrienLE