Don't Just Trust Your Memory: Evidence-Linked Memory Reduces Memory-Induced Degradation in LLM Agents
Across five agent benchmarks, WikiTrace beat a baseline given the same memory insights without their supporting evidence.
The system links each lesson to a summary of why the original task succeeded or failed, plus the full task trajectory. Agents can inspect those records when a lesson lacks detail, conflicts with another, or comes from a failed task. The authors report fewer cases where memory causes an agent to fail a task it could solve without memory. The paper is under review.
ArXiv · AI/CL/LG's note
The system links each lesson to a summary of why the original task succeeded or failed, plus the full task trajectory. Agents can inspect those records when a lesson lacks detail, conflicts with another, or comes from a failed task. The authors report fewer cases where memory causes an agent to fail a task it could solve without memory. The paper is under review.
ArXiv · AI/CL/LG's note
score 4