S-Curves and Saturating Growth

Definition

An S-curve is a trajectory with early slow growth, a middle phase of rapid increase, and a late phase that slows as it approaches a limit. A canonical saturating model is logistic growth:

$$\frac{dx}{dt}=rx\left(1-\frac{x}{K}\right),$$

where r is the intrinsic growth rate and K is the carrying capacity or saturation level.

See Logistic Sigmoid and Saturating Growth for the compact derivation and local-exponential comparison.

Why It Matters

S-curves are a guardrail against overreading short exponential-looking windows. In AI forecasting, benchmark progress, adoption, and capability narratives, a local trend can look exponential while the underlying process is bounded by data, compute, energy, evaluation headroom, or deployable use cases.

Formalism / Key Objects

• Logistic solution:

$$x(t)=\frac{K}{1+A{e}^{-rt}},$$

for a constant A set by the initial condition.

• Local exponential regime: when x \ll K, the term (1-x/K) is near 1, so logistic growth resembles dx/dt \approx r x.
• Saturation regime: as x approaches K, the growth term tends to zero.
• Stacked S-curves: a sequence of new capacities, technologies, or adoption pools can make many bounded processes look like persistent aggregate growth.

Connections

• Sigmoid Functions is the shape family behind the logistic and related curves.
• Logistic Sigmoid and Saturating Growth stores the reusable logistic equations.
• Double Descent is not a saturating growth curve, but it is another example where model-performance curves can violate simple monotone extrapolation.
• AI Evaluation and Benchmarking can use this idea when thinking about benchmark saturation and misleading progress curves.
• Deep Learning Fundamentals uses it as background for interpreting learning curves and AI-scaling claims.

Common Confusions

• An exponential fit over a short window does not by itself justify an unbounded exponential model.
• Not every S-shaped curve has the same causal mechanism; the capacity term must be tied to concrete limiting inputs.
• Stacked S-curves can mimic long-run exponential growth for a while, so the distinction is often empirical and causal rather than just visual.

Key Sources

• LessWrong2026 - All Exponentials are Eventually S-Curves
• Wikipedia2026 - Sigmoid Function