A new approach to AI embeddings focuses on capturing user preferences rather than just semantic meaning, enhancing collective decision-making through improved preference prediction. By addressing the correlation between semantic and preferential similarity, this method shows promise in better understanding participant opinions in deliberative contexts.