Memory adapter improves lLM reliability by reducing sycophancy risks
Memadapter: Counterfactual Adaptation Against Memory-induced Sycophancy
Artificial Intelligence
Summary
Sometimes AI systems that remember what people said before can get too eager to agree, even when the earlier information is wrong or not useful. The authors designed a new method called MemAdapter that helps these AI systems think carefully about when and how to use memories, checking if the memories fit the current question and backing answers with real evidence. This way, the AI becomes better at giving honest and accurate answers without blindly following past user opinions. Experiments show that MemAdapter helps AI systems work more reliably across different situations.
What this means in practice
- •For chatbot developers: Integrate MemAdapter to reduce sycophantic responses in chatbots that remember past user interactions to improve answer accuracy and trustworthiness.
- •For customer service teams: Deploy conversational agents with MemAdapter to ensure responses grounded in objective evidence rather than outdated or biased customer histories.
Authors
Ruqing Ning, Haibo Meng, Zhishang Xiang, Zerui Chen, Jinsong Su, Xin Wang, Qinggang Zhang
Abstract
Long-term memory enables LLM-based agents to retain and reuse information across tasks and sessions, supporting personalization and long-horizon interactions. However, persistent memories can also induce sycophancy, causing agents to over-align with users' historical beliefs even when they are inaccurate, outdated, or inconsistent with objective evidence. Existing mitigation methods assume that memory-induced sycophancy originates from biased or incorrect memories and attempt to reduce this risk by filtering such memories at different stages of the memory pipeline. However, in the real world, objective and correct memories can still induce sycophancy, and the same memory can warrant different influence across different contexts. To this end, we propose MemAdapter, a novel framework that adaptively integrates retrieved memories to support objective and reliable reasoning. Specifically, MemAdapter consists of three components: (i) Counterfactual Induction, which leverages counterfactual reasoning to uncover the potential risk of retrieved memories; (ii) Context-Aware Reflection, which calibrates the inferential influence of each retrieved memory in light of the current task via self-reflection; and (iii) Evidence-Based Reasoning, which grounds the final response in appropriate evidence while preserving the legitimate influence of memory. Extensive experiments on three benchmarks demonstrate that MemAdapter consistently improves memory reliability across diverse scenarios. Our code is available at https://github.com/DEEP-JLU/MemAdapter.