This paper explores the phenomenon of *emergent abilities* in large language models, which are abilities that are not present in smaller models but become apparent in larger ones. These abilities cannot be predicted by simply extrapolating the performance of smaller models and often show a qualitative change, known as a phase transition, at a critical threshold of scale. The authors discuss various tasks and techniques where emergent abilities have been observed, such as few-shot prompting, augmented prompting strategies, multi-step reasoning, instruction following, program execution, and model calibration. They also highlight the importance of understanding the factors that enable these abilities and the potential risks associated with them. The paper concludes by discussing future directions, including further model scaling, improved model architectures, data scaling, better techniques for prompting, frontier tasks, and a deeper understanding of emergence.This paper explores the phenomenon of *emergent abilities* in large language models, which are abilities that are not present in smaller models but become apparent in larger ones. These abilities cannot be predicted by simply extrapolating the performance of smaller models and often show a qualitative change, known as a phase transition, at a critical threshold of scale. The authors discuss various tasks and techniques where emergent abilities have been observed, such as few-shot prompting, augmented prompting strategies, multi-step reasoning, instruction following, program execution, and model calibration. They also highlight the importance of understanding the factors that enable these abilities and the potential risks associated with them. The paper concludes by discussing future directions, including further model scaling, improved model architectures, data scaling, better techniques for prompting, frontier tasks, and a deeper understanding of emergence.