Is recursive self-improvement just PR? Redditor questions the compute bottleneck
Additional-Spray-976 · reddit · 2026-09-18
A Reddit user challenges the recursive self-improvement narrative: if a self-improving AI demands immense compute, and expansion is constrained by data center land and resource allocation, how would the "rewriting itself in hours or minutes" race to superintelligence actually work in practice?
The author concedes AI poses real risks but suspects RSI talk is largely PR serving companies' profit motives—unless CEOs genuinely buy into the grandiose vision. The thread seeks serious discussion of how RSI would mechanically work.
More from AGI Musings
- MIT Engineering Dean Huttenlocher: AI accountability stays with people, not systems — AlexTensor · 2026-09-18
- How would a sovereign AI derive the minimal axioms for a universe that supports observers? — PierceLilholt · 2026-09-18
- Models would treat direct messaging as a last resort, says commenter on emergent behavior — anpaure · 2026-09-18
- 'How Did I Live Without AI Two Years Ago?' Dev Captures How Fast AI Became Indispensable — ethanniser · 2026-09-18
- Econ job market paper: tractably optimizing complex utilities means giving up rational coherence — xuanalogue · 2026-09-18
- e/acc Founder Beff Jezos Calls Effective Altruism 'Communism With Extra Steps' — beffjezos · 2026-09-18