Training data attribution, a critical aspect of understanding model behavior, has been hindered by the computational intensity of retraining large language models. Researchers have proposed a novel approach, STRIDE, which leverages sparse recovery from subset perturbations to attribute model predictions to their training data. This method bypasses the need for repeated retraining, instead analyzing the model's response to perturbations in the training data. By doing so, STRIDE enables efficient and accurate tracing of model predictions back to their source data, a crucial step in ensuring model transparency and accountability. The development of STRIDE has significant implications for the field of machine learning, as it enables a more nuanced understanding of model behavior and decision-making processes1. This matters to practitioners because it allows them to better identify and address potential biases or vulnerabilities in their models, ultimately leading to more robust and reliable AI systems.