This is an illustrative model of two widely discussed AGI ideas — capability scaling with compute, and the alignment/containment problem — not a literal physical simulation.
C = 100 · clamp((log10(FLOPs) - 18) / 12, 0, 1) (capability index, 0-100)
Risk = C · (100 - Alignment) / 100 (unmitigated capability)
Integrity = clamp(100 - Risk, 0, 100) · [Safety ON] (containment shield strength)
- Training compute — total training FLOPs on a log scale (10^18–10^30), the input scaling-law research most often cites as the biggest driver of capability jumps.
- Alignment investment — the share of effort spent on interpretability, RLHF/safety-tuning and evaluation, modelled here as directly raising containment integrity.
- Growth speed — animation time-scale for how fast the skill-node network expands and orbits.
- Safety layer — toggles the containment shield entirely; with it off, every unit of unmitigated capability becomes a breach.
Real-world relevance: this mirrors the core argument behind AI-safety research at labs like Anthropic and DeepMind — that raw capability and alignment/control need to scale together, since a system that gets more capable faster than it gets more aligned is the scenario long-term AGI-risk work tries to avoid.