This paper explores the potential of large language models (LLMs) as zero-shot anomaly detectors for time series data. The authors introduce SiGLLM, a framework that converts time series data into text format suitable for LLMs and presents two methods: PROMPTER and DETECTOR. PROMPTER uses prompts to elicit LLMs to identify anomalies, while DETECTOR leverages LLMs' forecasting capabilities to detect anomalies by comparing the original and forecasted signals. The study evaluates these methods on 11 datasets and finds that DETECTOR outperforms PROMPTER in terms of F1 score, achieving a significant improvement of 135%. However, LLMs still fall short of state-of-the-art deep learning models, which achieve 30% better performance. The paper also discusses the limitations and challenges of using LLMs for anomaly detection, including the need for context window size and the potential for false positives. Despite these limitations, the results suggest that LLMs have the potential to be effective zero-shot anomaly detectors for time series data.This paper explores the potential of large language models (LLMs) as zero-shot anomaly detectors for time series data. The authors introduce SiGLLM, a framework that converts time series data into text format suitable for LLMs and presents two methods: PROMPTER and DETECTOR. PROMPTER uses prompts to elicit LLMs to identify anomalies, while DETECTOR leverages LLMs' forecasting capabilities to detect anomalies by comparing the original and forecasted signals. The study evaluates these methods on 11 datasets and finds that DETECTOR outperforms PROMPTER in terms of F1 score, achieving a significant improvement of 135%. However, LLMs still fall short of state-of-the-art deep learning models, which achieve 30% better performance. The paper also discusses the limitations and challenges of using LLMs for anomaly detection, including the need for context window size and the potential for false positives. Despite these limitations, the results suggest that LLMs have the potential to be effective zero-shot anomaly detectors for time series data.