Back to feed
arXiv cs.AI·

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

Signal
72
Hype
18
In three linesCMI (Causal Memory Intervention) selects relevant memories for long-horizon LLM agents through controlled causal interventions rather than semantic similarity. Causal-LoCoMo benchmark introduced with useful memories, distractors, and synthetic harmful memories. CMI outperforms baselines (vector, graph, reflection, summary) in robustness against misleading memories.
Read source
Your take?
AI AgentsReasoningBenchmarksPapers

Summary generated by Claude — human-verified