Optimization Compounds

December 4, 2024

Alternative titles: there is always a bottleneck; iterated Amdahl's law; Amdahl's job security.

Optimization diagram
If the speed of a system doubles, so does the impact of remaining optimizations.

Due to Jevons paradox, increases in efficiency induce corresponding increases in demand. Thus if an optimization today would save $X, after the efficiency of the broader system doubles we expect the optimization to save more than $2X. At Suno demand for the latest models is essentially infinite. The lower the cost, the more applications we find.