This paper investigates the effectiveness of large language models (LLMs) for identifying gravitational wave signals in complex, noisy observational data, demonstrating superior performance over traditional neural networks with significantly fewer labeled samples and without reliance on large simulated datasets.