Narrow ASI Inevitable, Broad ASI Not: Ramez and Sayashk Debate Task-Level Limits
sudoraohacker · x · 2026-10-11
- Ramez Naim's take: AI can become superhuman purely via compute in highly verifiable domains like math and cybersecurity, making "narrow ASI" seem inevitable. But most important domains face fundamental limits or real-world data/experimentation constraints — "broad, general ASI" is far from certain.
- Sayash Kapoor's framework: This is the key crux separating the "normal technology" worldview from the "superintelligence" worldview — and it has nothing to do with the pace of progress, only assumptions about the world. Different tasks have different inherent limits: some are like "making a building taller" (no ceiling), others hit hard physical or empirical walls. Superhuman cybersecurity and math performance won't automatically transfer to many consequential tasks.
- Both converge: the real question isn't speed but whether unbreakable ceilings exist per task.
More from AGI Musings
- 2026 is everyone trying to make themselves illegible to AI and legible to capital — kevinakwok · 2026-10-11
- X's Nikita Bier: Welfare Applicants Should Lose Benefits for Not Using AI to Make Money — nikitabier · 2026-10-11
- Diamandis: Musk believes in universal high income, with healthcare better than today's billionaires' — PeterDiamandis · 2026-10-11
- AI's underappreciated virtue: filling knowledge gaps without judgment — arjunrajlab · 2026-10-11
- Musk says humanity is a bootloader for AI; critics argue Altman's merge thesis is equally 'extinctionist' — PolarBearby · 2026-10-11
- When AI writes and reviews the code, who still understands the codebase? — KL_AIC · 2026-10-11