Time horizon

Aligned superintelligence arrives before lock-in windows close and humanity can credibly commit to reflect rather than act.

Current scaling approaches hit a wall before producing transformative AI, so strategy selection is premature because the forecast capability will not arrive.

Something will emerge; specific interventions are more likely wrong than right, so staying uncommitted preserves option value.

Harms from lower capability AI are informative about harms from higher capability AI, and deployment feedback outperforms fast scaling.