STELLA proposes a method to guide Large Language Models for improved time series forecasting by employing semantic abstractions, potentially leading to more accurate and interpretable predictions in various real-world applications.