Fast vs slow takeoff

The fault line is about time. One side reads recent capability scaling — Claude 4 series, multimodal reasoning, agentic harnesses — as evidence the curve is bending sharply, with each generation buying less external alignment runway than the last. The other side reads the same evidence as gradual: capabilities arrive, are deployed, are observed, and the operating envelope adapts. Disagreements about scheduling cascade into disagreements about which interventions are worth funding now vs. later.

What’s in scope for this cluster: positions on capability-curve shape, on the breadth of compute overhang, on whether interpretability buys time or merely reveals what’s already happened, and on the operational difference between a 6-month and a 6-year transition window.

What’s out of scope: the orthogonal “should we build it at all” debate. That’s a different cluster.