Large language models and machine learning models are built on statistical patterns in historical data and often fail in the face of structural break events such as the COVID-19 liquidity crisis in March 2020 and the UK pension crisis in 2022.