LLM在记忆系统中的位置决定遗忘能力，不同配置各有优劣。
Control-Plane Placement Shapes Forgetting: An Architectural Study of Agent Memory Across Thirteen System Configurations
- 通过13种配置对比，研究LLM在记忆管道中的位置影响
- 控制平面时机的LLM可实现93.2%整体遗忘准确率，提升27.8个百分点
- 适合关注记忆删除精度与系统架构设计的研究者
LLM在智能体记忆系统中的位置——位于召回层（检索存储事实）与控制层（通过覆盖、释放、清除等操作修改事实）之间——决定了系统能应对哪些遗忘失败模式。在385个对抗性案例上比较13种系统配置，发现三种放置策略具有部分互补性：确定性原语对词汇/时间类任务足够，但无法处理归一化（标识符混淆下5%，跨语言场景0%）；插入时使用LLM可恢复归一化（100%），但无助于意图感知删除（前缀冲突和复合事实场景为0%）；在突变时引入钩子可实现意图感知删除（78-85%），同时显著提升几乎所有类别性能（总体91.7-93.2%），每次运行成本0.17美元，单次突变延迟2.3秒（对比确定性方法64-191毫秒，召回路径不变）。通过ForgetEval评估框架（1000个模板案例+385个对抗案例，含132个手工构造与253个LLM生成的验证案例）结合六种方法的Adapter协议，以子串匹配评分并支持异构记忆存储接入（130行代码），经10名标注者评估（Fleiss' kappa = 0.958）及77个外部作者案例验证，确认归一化不对称性并放大联合部署增益（+27.8点）。生产故障主要源于遗忘而非召回失败，但现有基准仅衡量召回。ForgetEval与所有适配器已开源（MIT许可）。
原文摘要 · Abstract (English)
Where an LLM sits in an agent memory pipeline -- between the recall plane that retrieves stored facts (extensively benchmarked) and the control plane that mutates them via supersede, release, purge (largely untested) -- shapes which forgetting failure modes the system recovers. Comparing thirteen system configurations on a 385-case adversarial surface, we observe three placement regimes with partly complementary coverage: deterministic primitives suffice for lexical/temporal categories but fail canonicalization (5% on identifier-obfuscation, 0% on cross-lingual); inscribe-time LLM recovers canonicalization (100%) but cannot help intent-aware deletion (0% on prefix-collision and compound-fact); a mutation-time hook recovers intent-aware deletion (78-85%) and brightens nearly all categories simultaneously (91.7-93.2% overall, $0.17 per 385-case run, 2.3s/case mutation latency vs. 64-191ms/case deterministic, recall path unchanged). We expose the trade-off via ForgetEval, a 1000-case templated suite plus a 385-case adversarial layer (132 hand-crafted + 253 LLM-drafted oracle-validated) scored by deterministic substring match, paired with a six-method Adapter Protocol with honest N/A scoring that lets heterogeneous memory stores enter in 130 lines. Admission is corroborated by 10-annotator IAA (Fleiss' kappa = 0.958) and a 77-case external-authored subset (four blind contributors) that replicates the canonicalization asymmetry and amplifies the joint-placement lift (+27.8 pt). Production failures are predominantly forgetting failures rather than recall failures, yet existing benchmarks measure only recall. ForgetEval and all adapters are released under MIT.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。