Search papers, labs, and topics across Lattice.
This study investigates the memory-clarification boundary in LLM agents, focusing on whether interaction-derived information should be persisted, verified, or clarified with users. Using a dataset of 140 scenarios, the authors find that models like Claude and Qwen are more reliable at verifying changing facts than at asking users for clarification, with few-shot prompting significantly improving accuracy. The research highlights that while a policy prompt can reduce erroneous memory persistence, the overall accuracy gains remain modest, emphasizing the complexity of memory management in LLMs.
Models can verify facts more reliably than they can clarify ambiguities, raising questions about how we design memory systems in LLMs.
Persistent memory can personalize an LLM agent, but an incorrect durable update can silently distort future behavior. We study the memory-clarification boundary: whether interaction-derived information should be persisted, used only in the current context, re-verified, or clarified with the user. MCB contains 140 primary scenarios, split into 70 development and 70 held-out items, plus a separate 70-item contrast set. It evaluates both action labels and structured tool-call selection. Two non-authors independently label the 70 held-out primary and 70 contrast items (97.1% agreement, Cohen's kappa = 0.962); a blind third resolves four disagreements, replacing eight author labels by non-author majority. Across Claude and Qwen, models verify changing facts more reliably than they ask users to resolve ambiguity. Bare Qwen asks on 0/12 clarification items while verifying 12/18 freshness items. Few-shot prompting raises accuracy from 0.557 to 0.771 (paired delta = +0.214, Holm-adjusted exact McNemar p_H = 0.002), yet clarification recall remains 0.333. The policy prompt reduces erroneous persistence from 0.243 to 0.100 (p_H = 0.038), although its accuracy gain is not significant. Label-tool agreement is 57% for each Claude model and 23% for Qwen; Qwen accuracy falls from 0.557 to 0.343 (p_H = 0.047). Memory evaluation must test both stated decisions and tool-call choices.
