How Accurate Have AI Progress Forecasts Been So Far?

The calibration angle is useful: separating capability forecasts from deployment and adoption forecasts makes misses easier to diagnose.

I think this will widen the gap even more.