Do AIs face inherent task limits? Chess example tests superintelligence worldview
gleech · x · 2026-10-14
SayashK argues the key crux separating the "normal technology" and "superintelligence" worldviews is whether tasks have different inherent limits: some tasks (taller buildings) have no ceiling, while others do. He uses chess—an in-principle solvable game where humans sit far from the limit—as an intuition pump that even Category 2 tasks can see huge AI overperformance. Peter Salib counters that this example actually undermines the framework: since humans are nowhere near chess's optimum, AI can do vastly better, so the category labels don't predict much.
More from AGI Musings
- Frontier labs reject free labor: GPU costs dwarf salaries as LLM work goes 'manufacturing-like' — dotey · 2026-10-14
- 92% of UK employers report candidate misconduct as AI cheating tops job interviews — nordicinst · 2026-10-14
- Google Fellow John Platt: AI rerouting cut flight contrails by 62% — ziv_ravid · 2026-10-14
- Split your computers into high/low-risk zones before AI agents go autonomous — onektruths · 2026-10-14
- Reading books while your agents work: AI is giving users their attention back — martyamark · 2026-10-14
- Cerebras CEO Andrew Feldman tells AI doomsayers to 'maybe stop talking' — TechCrunch AI · 2026-10-14