Researchers have introduced Long-Term Embeddings (LTE) to address the limitations of modern transformer-based sequential recommenders, which often prioritize short-term intent over stable long-term preferences. By serving as a high-inertia contextual anchor, LTE helps to mitigate recency bias and capture enduring user preferences. Unlike simply extending sequence lengths, which can be computationally inefficient and dominated by recent interactions, LTE provides a more balanced approach to personalization. This innovation has significant implications for sequential recommendation systems, enabling them to better account for long-term user interests and behaviors. The development of LTE is particularly noteworthy in the context of state-aligned activity involving transformer technology, which shifts the threat model from criminal to geopolitical1. So what matters to practitioners is that LTE can enhance the accuracy and robustness of recommendation systems, ultimately leading to more effective and personalized user experiences.