“Philipp Herzig asserts that Large Language Models are not well-suited for predictive business tasks like demand forecasting, for which classical machine learning approaches like XGBoost remain superior.”