发现大模型会偏好过时信息，提出新方法精准修复。
Right Knowledge, Wrong Answer: Characterizing Parametric Temporal Conflict in Open-Weight Language Models
- 构建了8746个时间冲突数据集，量化模型对新旧信息的偏好。
- 用日期前缀提示可恢复61%-81%的新信息，激活修补成功率72%-85%。
- 揭示模型上层存在定向表示，适合研究模型推理偏差者参考。
语言模型可能同时编码过时事实及其更新版本。本文提出参数化时间冲突（PTC），即新事实虽可恢复但默认前向传播仍偏好旧事实。我们构建了一个包含8,746个Wikidata职位继承关系的确定性验证基准，评估了四个开源语言模型在三个系列中的表现。使用日期前缀提示可在61%-81%的PTC案例中恢复新事实；激活修补可使72%-85%的预测发生翻转，并定位到特定模型的上层区域。残差流引导优于归一化随机方向，表明存在方向特异性表示。结果表明PTC反映的是局部表征偏好而非知识缺失。恢复效果基于人工识别的冲突进行评估，因自动检测尚不可靠。我们已公开该基准、代码与统计结果。
原文摘要 · Abstract (English)
Language models may encode both outdated facts and their newer replacements. We introduce Parametric Temporal Conflict (PTC), where the newer fact is present and recoverable, but the default forward pass prefers the outdated one. We release a deterministically verified benchmark of 8,746 Wikidata position-holder transitions and evaluate four open-weight language models across three families. A date-prefix prompt recovers the newer fact in 61-81% of PTC cases. Activation patching flips predictions in 72-85% of cases and localizes the preference to model-specific upper-layer regions. Residual-stream steering outperforms norm-matched random directions, indicating direction-specific representations. These results show that PTC reflects a localized representational preference rather than missing knowledge. Recovery is measured on oracle-identified conflicts because automatic detection remains unreliable. We release the benchmark, code, and statistics.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。