Does Exponential Compute Mean Exponential Intelligence? Tom Davidson and Ramez Clash Over Intelligence Explosion
connoraxiotes · x · 2026-09-29
A substantive debate between researcher Tom Davidson and Ramez Naim on whether growing compute leads to an intelligence explosion.
- Ramez's objection: Epoch's r>1 evidence covers computational efficiency, not intelligence; compute-to-capability scaling is steeply sub-linear, so the feedback loop via more agents would see roughly logarithmic diminishing returns.
- Davidson's rebuttal: The common claim that exponentially growing compute yields only linearly growing intelligence rests on vibes-based intelligence measurements — e.g., equating GPT-2→GPT-3 with GPT-3→GPT-4, or treating each 10% benchmark gain as equivalent.
- He argues the better yardstick is AI's economic value-add: as training compute grew exponentially, so have AI revenues — implying exponential compute yields exponential intelligence. Even impoverished RSI models that only count efficiency improvements predict an intelligence explosion, he notes.
More from AGI Musings
- AI safety researcher floguo reflects: vague ideals of 'human flourishing' need firmer foundations — floguo · 2026-09-29
- Nathan Lambert: it's a very hard time to think and criticize in public in AI — natolambert · 2026-09-29
- a16z podcast: personal AI agents are exploding, and the best ones will be invisible — a16z · 2026-09-29
- Findings of ACL already separates papers from talks, so 'AI-authored' rules are moot, scholar argues — ipeirotis · 2026-09-29
- Humanities will rule the future of higher education, scholar argues in AI era — begusgasper · 2026-09-29
- Education writer suggests keeping young kids away from AI for now — benjaminjriley · 2026-09-29