The Anatomy of an Edit: Mechanism-Guided Activation Steering for Knowledge Editing
For researchers in knowledge editing, this work provides a mechanistic understanding of edit implementation and a new method that improves editing reliability without weight modification.
The paper investigates how knowledge edits are implemented inside LLMs by using post-edit neuron-level attribution, revealing that mid-to-late attention promotes the new target while attention and FFN modules cooperate to suppress the original fact. Based on these findings, they propose MEGA, a mechanism-guided activation steering method that achieves strong editing performance on CounterFact and Popular datasets for GPT2-XL and LLaMA2-7B.
Large language models (LLMs) are increasingly used as knowledge bases, but keeping them up to date requires targeted knowledge editing (KE). However, it remains unclear how edits are implemented inside the model once applied. In this work, we take a mechanistic view of KE using neuron-level knowledge attribution (NLKA). Unlike prior work that focuses on pre-edit causal tracing and localization, we use post-edit attribution -- contrasting successful and failed edits -- to isolate the computations that shift when an edit succeeds. Across representative KE methods, we find a consistent pattern: mid-to-late attention predominantly promotes the new target, while attention and FFN modules cooperate to suppress the original fact. Motivated by these findings, we propose MEGA, a MEchanism-Guided Activation steering method that performs attention-residual interventions in attribution-aligned regions without modifying model weights. On CounterFact and Popular, MEGA achieves strong editing performance across KE metrics on GPT2-XL and LLaMA2-7B. Overall, our results elevate post-edit attribution from analysis to engineering signal: by pinpointing where and how edits take hold, it powers MEGA to deliver reliable, architecture-agnostic knowledge edits.