Know It, Act on It: Investigating Memory Utilization in LLM Personalization
Source: https://arxiv.org/abs/2607.29433
Authors: Zhaoxin Feng, Jianfei Ma, Emmanuele Chersoni
Published: 2026-07-31
Categories: cs.CL
Grounding: opencli arxiv metadata + Obsidian paper digest excerpt (OpenClaw定时任务/论文流水线/2026-08-04-论文流水线.md)
摘要（中文）
论文区分模型知道用户偏好和模型按偏好行动作者设计 Know / Act paired tests，在 16 个系统5 种 memory architectures1,000 个偏好上测试摘要指出 Agent 经常能通过 recall 测试，却没有在行为场景中体现同一偏好；医疗和心理相关偏好上的利用弱点尤其明显
Abstract
As large language model (LLM) agents evolve into personalized companions, memory has emerged as a core capability. However, LLMs face a knowledge utilization problem: they may fail to act on relevant user preferences even when they are fully present in context. When an agent fails to tailor its response in a context where previously shared user preferences should matter, it is unclear whether the model failed to remember that information or remembered it but failed to use it. To isolate this breakdown, we introduce a decoupled evaluation paradigm that administers paired Know and Act tests to the same user preference. We conduct large-scale experiments across 16 systems and five memory architectures, evaluating 1,000 preferences embedded at three levels of expression strength.
Intake rationale
- Category: agents
- Quality score: 4/5
- Why it matters: Know It Act on It separates remembering a preference from actually using it in behavior.
Obsidian evidence excerpt
论文区分“模型知道用户偏好”和“模型按偏好行动”。作者设计 Know / Act paired tests，在 16 个系统、5 种 memory architectures、1,000 个偏好上测试。摘要指出 Agent 经常能通过 recall 测试，却没有在行为场景中体现同一偏好；医疗和心理相关偏好上的利用弱点尤其明显。