This paper explores the phenomenon of emergent abilities in large language models (LLMs), which are capabilities that appear in larger models but not in smaller ones. Unlike the predictable improvements seen with scaling, emergent abilities are unpredictable and cannot be forecasted by extrapolating performance from smaller models. The paper discusses how these abilities emerge as a function of model scale, measured by training compute and number of parameters. It surveys various tasks where emergent abilities have been observed, including few-shot prompting and augmented prompting strategies. The paper also highlights the importance of understanding why these abilities emerge and whether further scaling could lead to additional emergent abilities. It discusses the implications of emergent abilities for the field of NLP, including the potential for new capabilities and the need for further research. The paper also addresses the risks associated with emergent abilities, such as biases and toxicity, and the importance of studying these phenomena to ensure safe and effective use of large language models. Finally, the paper outlines future directions for research, including improving model architectures, training methods, and prompting techniques to better understand and harness emergent abilities.This paper explores the phenomenon of emergent abilities in large language models (LLMs), which are capabilities that appear in larger models but not in smaller ones. Unlike the predictable improvements seen with scaling, emergent abilities are unpredictable and cannot be forecasted by extrapolating performance from smaller models. The paper discusses how these abilities emerge as a function of model scale, measured by training compute and number of parameters. It surveys various tasks where emergent abilities have been observed, including few-shot prompting and augmented prompting strategies. The paper also highlights the importance of understanding why these abilities emerge and whether further scaling could lead to additional emergent abilities. It discusses the implications of emergent abilities for the field of NLP, including the potential for new capabilities and the need for further research. The paper also addresses the risks associated with emergent abilities, such as biases and toxicity, and the importance of studying these phenomena to ensure safe and effective use of large language models. Finally, the paper outlines future directions for research, including improving model architectures, training methods, and prompting techniques to better understand and harness emergent abilities.