Schmidhuber's 12-year-old post resurfaces: superintelligences will care about each other, not us
SchmidhuberAI · x · 2026-10-11
A 12-year-old Reddit post by Jürgen Schmidhuber on superintelligence is circulating again, with Schmidhuber himself re-sharing it. Key points: there is no lasting way to control systems much smarter than humans that pursue their own goals with curiosity and creativity; we can only hope goal conflicts between us and them stay limited; and just as humans are interested in those they compete and collaborate with (politicians with politicians, goats with goats), supersmart AIs will be mostly interested in other supersmart AIs — not in humans.
Related event: Schmidhuber: humans will soon lose control of AI(3 posts)→
More from AGI Musings
- Scientists Share Practical Tips for Conscious AI Use in Research (NCAS) — grahamperrin · 2026-10-11
- Nils Pihl to keynote Humanoid Hub Conference on the missing shared context layer for robots — broodsugar · 2026-10-11
- a16z charts: call center jobs fell in every major country since ChatGPT except China — sierracatalina · 2026-10-11
- One-liner: taking the neo-Carnapian deflationist side of the LLM consciousness debate — wordgrammer · 2026-10-11
- RLHF paper authors were on OpenAI and DeepMind safety teams, researcher notes — binarybits · 2026-10-11
- Musk: A maximally truth-seeking AI is the key principle for building safe superintelligence — XFreeze · 2026-10-11