CLaRE-ty Amid Chaos: Quantifying Representational Entanglement to Predict Ripple Effects in LLM Editing

2026-03-11

AI Summary

Editing large language models often triggers unpredictable ripple effects—unintended behavioral changes across unrelated tasks. To address this, this work proposes CLaRE, the first lightweight, gradient-free method that quantifies entanglement among factual representations using forward activations from a single layer, enabling the construction of large-scale factual entanglement graphs. CLaRE facilitates efficient creation of edit-protection sets, audit trails, and red-teaming evaluations. Experimental results demonstrate that CLaRE achieves an average 62.2% improvement in Spearman correlation, accelerates inference by 2.74×, reduces peak GPU memory usage by 2.85×, and substantially lowers storage overhead compared to existing approaches.

Abstract

The static knowledge representations of large language models (LLMs) inevitably become outdated or incorrect over time. While model-editing techniques offer a promising solution by modifying a model's factual associations, they often produce unpredictable ripple effects, which are unintended behavioral changes that propagate even to the hidden space. In this work, we introduce CLaRE, a lightweight representation-level technique to identify where these ripple effects may occur. Unlike prior gradient-based methods, CLaRE quantifies entanglement between facts using forward activations from a single intermediate layer, avoiding costly backward passes. To enable systematic study, we prepare and analyse a corpus of 11,427 facts drawn from three existing datasets. Using CLaRE, we compute large-scale entanglement graphs of this corpus for multiple models, capturing how local edits propagate through representational space. These graphs enable stronger preservation sets for model editing, audit trails, efficient red-teaming, and scalable post-edit evaluation. In comparison to baselines, CLaRE achieves an average of 62.2% improvement in Spearman correlation with ripple effects while being $2.74\times$ faster, and using $2.85\times$ less peak GPU memory. Besides, CLaRE requires only a fraction of the storage needed by the baselines to compute and preserve fact representations. Our entanglement graphs and corpus are available at https://anonymous.4open.science/r/CLaRE-488E.

Problem

• large language models
• model editing
• ripple effects
• representational entanglement
• knowledge updating

Innovation

• representational entanglement
• model editing
• ripple effects
• forward activations
• entanglement graph