The Unpredictable Abilities Emerging From Large AI Models (via) Nice write-up of the most interesting aspect of large language models: the fact that they gain emergent abilities at certain “breakthrough” size points, and no-one is entirely sure they understand why.