Schmidhuber's AI Blog Surveys Recursive Self-Improvement from 1987 and the 1991 Roots of Today's AI Boom
SchmidhuberAI · x · 2026-09-20
Jürgen Schmidhuber points to his AI Blog as the best overview of his work, with two highlighted pieces:
- "Recursive Self-Improvement (RSI) since 1987": with Anthropic/OpenAI/Sakana AI and RSI-branded startups all discussing recursive self-improvement, he notes he published the first concrete RSI algorithms in 1987 when compute was 100M times more expensive. The survey covers meta-evolution since 1987, gradient-descent-based RSI since 1992, self-modifying policies since 1994, asymptotically optimal RSI for curriculum learning since 2002, the Gödel Machine since 2003, and RSI combined with artificial curiosity. He argues software RSI is now practical, but full RSI requires self-improving hardware too.
- "Munich 1991: the Roots of the Current AI Boom" (preface by David Ha): his team's 1991 publications including the first kind of Transformer (unnormalized linear Transformer) and pre-training & neural net distillation.
More from AGI Musings
- When a Prompt Yields Publishable Work, How Should Journals Respond? — JJitsev · 2026-09-20
- Daniel Rock: humans are jagged too — a calculator would be shocked at our arithmetic — danielrock · 2026-09-20
- Worried About Rogue AI but Not Mountain Lions Stalking San Francisco? — omooretweets · 2026-09-20
- Daniel Rock: Humans are jagged too — a calculator would be shocked at our arithmetic — danielrock · 2026-09-20
- Alignment researcher: short-timeline arguments lack mechanistic rigor, risking misdirected AI safety work — JacquesThibs · 2026-09-20
- Are AI agents changing how engineers think, not just how fast they ship? — Relevant-Potential17 · 2026-09-20