PRISM Edit：适用于所有时间答案的单一向量
PRISM Edit: One Vector for All Temporal Answers
浏览论文内容
中文总结 AI 辅助
研究针对大语言模型时间事实更新问题，基于因果追踪发现其内部计算支持新旧答案区分，进而引入PRISM Edit，通过优化单一多义词表示及利用固有调制路径，在新基准上评估，相比基线提升了时间一致性等指标且速度更快。
中文摘要 AI 辅助
模型编辑可让大语言模型（LLMs）无需重新训练就能保持更新，但时间事实揭示了当前定位与编辑范式的局限性：更新并不总是替换。当事实发生变化时，新答案应成为当前答案，而旧答案在历史时间背景下可能仍然正确。基于此，我们用因果追踪表明LLMs已通过两阶段内部计算支持这种区分：早期MLP层检索与时间无关的主题表示，后期层用时间上下文对其进行调制以产生时间正确的答案。受此发现启发，我们引入PRISM Edit，它在不修改架构的情况下，跨时间上下文优化单个多义词表示，并利用模型固有的调制路径将其引导至时间正确的预测。我们在新引入的时间编辑基准TimeConflict和时间增强的CounterFact上进行评估。PRISM Edit平均比最佳基线提高了23.3的时间一致性（TC）和33.7的当前相对时间得分（CRS），同时速度快2倍以上。代码和数据可在指定网址公开获取。
英文摘要
Model editing keeps large language models (LLMs) up to date without retraining, but temporal facts expose a limitation of the prevailing locate-and-edit paradigm: an update is not always a replacement. When a fact changes, the new answer should become current while the old answer may remain correct in historical time contexts. Building on this insight, we use causal tracing to show that LLMs already support this distinction via a two-stage internal computation: early MLP layers retrieve a time-agnostic subject representation, and later layers modulate it with temporal context to yield the time-correct answer. Motivated by this finding, we introduce PRISM Edit, which optimizes a single polysemous representation across temporal contexts and leverages the model's inherent modulation pathway to route it to temporally correct predictions without requiring any architectural modification. We evaluate on TimeConflict, a newly introduced temporal editing benchmark, and on temporally augmented CounterFact. PRISM Edit improves multiple core metrics over the best baseline, most notably +23.3 Temporal Consistency (TC) and +33.7 Current Relative-time Score (CRS) on LLaMA-3, while being more than 2x faster. Code and data are publicly available at https://github.com/CheerCHuang/PRISM-Edit.
发表机构
- Tsinghua University（清华大学）
- ByteDance（字节跳动）
机构由 AI 辅助整理，请以论文原文为准。
