A new AI system called TRACE-Memory aims to improve personalized generation by selectively using user history. It does this in two stages: first, it identifies missing information from the public context and retrieves a pool of relevant evidence; second, it admits only the most useful evidence units to the model. The authors claim that their system outperforms other methods on various tasks, including text summarization and generation. --- Why it matters: This matters because current personalized generation systems often rely too heavily on user history, which can be irrelevant or redundant. TRACE-Memory's selective approach could lead to more accurate and efficient models in applications like chatbots and content recommendation systems. Source: https://arxiv.org/abs/2608.08446