← Back to Home

Marginal Returns to Intelligence

Why more intelligence doesn't always help proportionally

About This Simulator

Amodei identifies five factors that limit how much raw intelligence can accelerate progress. Even a "country of geniuses in a datacenter" faces constraints: experiments still take real time, some problems need new data that can't be generated instantly, some systems are intrinsically complex, humans may resist or cannot be removed from the loop, and physical laws impose hard limits.

Explore how these bottlenecks vary across domains. As you increase AI intelligence, watch how some fields scale beautifully while others hit walls — the key insight behind "marginal returns to intelligence."

1x 100x
1x
3x
Drug Discovery
Active Domain
1x
Intelligence Level
--
Overall Effectiveness
--
Primary Bottleneck

Bottleneck Analysis

Intelligence Effectiveness

How much of the intelligence increase translates to actual speedup?

Low (bottlenecked) Moderate High (scales well)
Key Insight

Select a domain and adjust intelligence to see how bottlenecks shape AI effectiveness.

Activity Log

[--:--:--] Simulator ready. Select a domain and adjust intelligence level.