Causal Intervention-Based Memory Selection for Long-Horizon LLM Agents

In three linesCMI (Causal Memory Intervention) selects relevant memories for long-horizon LLM agents through controlled causal interventions, rather than semantic similarity. A causally annotated benchmark (Causal-LoCoMo) evaluates robustness against misleading memories. CMI outperforms vector, graph, reflection, and full-history baselines.

Summary generated by Claude — human-verified