Large Language Models’ Emergent Abilities Are a Mirage

A new study suggests that sudden jumps in LLMs’ abilities are neither surprising nor unpredictable, but are actually the consequence of how we measure ability in AI.