Researchers have developed a new framework called LifeMem to improve the performance of large language models (LLMs) in social simulation. Current methods often rely on static agent profiles and can perpetuate identity essentialism, where demographic labels are used to homogenize responses within groups. LifeMem combines structured life-event retrieval with parametric memory for experience integration, allowing LLMs to better capture human-like diversity and respond more accurately to different scenarios. --- Why it matters: This matters because it addresses a limitation in current social simulation methods that can lead to biased and unrealistic models. Improving the performance of LLMs in this area can have significant implications for applications such as education, healthcare, and policy-making. Source: https://arxiv.org/abs/2608.19621