When Errors Become Memories: Causal Pathway Tracing in Multi-Turn Memory-Augmented LLMs
A structural causal model (SCM)-based framework for cross-turn error propagation in memory-augmented LLMs is proposed, and experiments show that error influence generally decays with interaction distance, while the memory-update pathway contributes more persistent effects than question feedback.
Shu-Yao Xiao, Sheng-Ling Wang, Xuan Chen et al.
· 0 citations