Schmidhuber: I published the first concrete RSI algorithms back in 1987
SchmidhuberAI · x · 2026-09-22
Amid today's buzz about Recursive Self-Improvement (RSI), Jürgen Schmidhuber recaps his decades-long track record: the first concrete RSI algorithms in 1987 (when compute was 100M× more expensive), gradient descent-based RSI in neural nets since 1992, self-modifying policies since 1994, asymptotically optimal RSI for curriculum learning since 2002, the mathematically optimal self-referential Gödel Machine since 2003, and RSI combined with artificial curiosity since 1990/1997. He argues cheap compute is now making RSI drive both software and physical AI, citing technical note IDSIA-9-26.
More from AGI Musings
- Tesla FSD vet: chasing bar charts has hollowed out LLM development — yunta_tsai · 2026-09-22
- Reddit debates an unusual AI trend chart: where does this curve end up? — wrcromagnum · 2026-09-22
- In the AI era, famous actors are locked out of creation while amateurs run free — Kyrannio · 2026-09-22
- 'Doom is the only forecast that pays either way': the warped incentives of AI doomerism — AlexTensor · 2026-09-22
- Gergely Orosz: most people won't hand digital wallets to AI shopping agents — HildeKuehne · 2026-09-22
- Robot arm grasps cube 71 minutes in: model writes its own kinematics, no training — Scobleizer · 2026-09-22