paper preprint

Emergent Abilities of Large Language Models

The paper documented benchmark abilities that appeared only above model-scale thresholds and argued that some prompted capabilities were not predictable by smooth extrapolation from smaller models.