How it works
Each glowing node represents a capability domain the system has bootstrapped for itself. The population of active nodes N follows a logistic recursive-self-improvement model:
dN/dt = r · N · (1 − N / C)
r = α − β
where α is the self-improvement rate (how fast the system rewrites and upgrades itself — the "intelligence explosion" term ΔIntelligence = α·Intelligence), β is the braking force from alignment and control measures, and C is the compute ceiling (hardware/data budget).
- If α ≤ β, growth is contained — N settles well below the ceiling (blue, calm).
- If α > β, capability races toward C — the node cloud fills and presses against the containment shell (red, unstable).
- Restart resets N to its seed value; Pause freezes the model to inspect a state.
Real-world echo: this is the same shape as debates over AI scaling laws — capability (compute × data × algorithms) racing against the maturity of alignment techniques.