Effortless and accurate time series analysis in astronomy using Large Language Models

The application of machine learning techniques in astronomy has been rapidly growing, addressing various challenges such as predicting orbital stability, classifying celestial objects, and analysing images. However, the emerging trend of using large language models (LLMs) presents a novel approach that relies on natural language processing and explicit task definitions, rather than on statistical algorithms or probabilistic models. In this talk, I'll demonstrate the exceptional capabilities of LLMs, specifically GPT-4 as well as some other open source alternatives, in analyzing visual patterns and accurately classifying asteroids as resonant or non-resonant without any training, fine-tuning, or coding beyond writing an appropriate prompt in natural language. By leveraging the power of LLMs, it is possible to achieve an accuracy, precision, and recall of 100% in differentiating between pure libration, circulation, and mixed (transient) cases of resonant angles.

This approach introduces a new paradigm in astronomical data analysis, where complex tasks requiring human expertise can be completed with minimal effort and resources. The implications of this study extend beyond the identification of mean-motion resonances, as the methodology can be applied to a wide range of astronomical problems that involve pattern recognition, outlier detection, and decision-making tasks. This presentation will discuss the experimental design, results, and potential applications of LLMs in astronomy, highlighting the significance of this innovative approach in advancing astronomical research.