MemAdapter uses counterfactual reasoning to curb memory-induced sycophancy in LLM agents
Ruqing Ning · hf · 2026-10-06
MemAdapter is a framework for memory-induced sycophancy in LLM agents, combining counterfactual induction to surface retrieval risks, context-aware reflection to calibrate each memory's influence, and evidence-based reasoning. It consistently improves memory reliability across three benchmarks. Code: github.com/DEEP-JLU/MemAdapter.
More from Research
- One-Sentence Tool Refusal Boosts Search Agent Abstention from 23% to 97% — _reachsumit · 2026-10-06
- CutBCE: TPU Kernel Eliminates OOM in Large-Vocabulary Recommendation Training, 91.9% Faster — _reachsumit · 2026-10-06
- Airbnb Trains RL Query Suggestion Model Using Search Reranker Scores as Reward — _reachsumit · 2026-10-06
- NVIDIA Study: Agentic Retrieval Boosts nDCG@10 by 8.7 Points but Takes 160x Longer — _reachsumit · 2026-10-06
- T-Search: open agentic retriever hits 61.3 Recall@10, beats larger open models — _reachsumit · 2026-10-06
- Qwen researcher to present QED-Nano work at COLM, Oct 5–9 — _lewtun · 2026-10-06