Competition of Mechanisms: Tracing How Language Models Handle Facts and Counterfactuals
This work addresses interpretability for AI researchers by moving beyond single-mechanism analysis, though it is incremental in extending existing methods to study interplay.
The paper tackles the problem of understanding how large language models handle conflicting factual and counterfactual information by proposing a 'competition of mechanisms' framework that traces multiple interacting mechanisms, revealing traces of competition across model components and attention positions that control mechanism strength.
Interpretability research aims to bridge the gap between empirical success and our scientific understanding of the inner workings of large language models (LLMs). However, most existing research focuses on analyzing a single mechanism, such as how models copy or recall factual knowledge. In this work, we propose a formulation of competition of mechanisms, which focuses on the interplay of multiple mechanisms instead of individual mechanisms and traces how one of them becomes dominant in the final prediction. We uncover how and where mechanisms compete within LLMs using two interpretability methods: logit inspection and attention modification. Our findings show traces of the mechanisms and their competition across various model components and reveal attention positions that effectively control the strength of certain mechanisms. Code: https://github.com/francescortu/comp-mech. Data: https://huggingface.co/datasets/francescortu/comp-mech.