Position Paper: Toward New Frameworks for Studying Model Representations

Mechanistic interpretability (MI) aims to understand AI models by reverse-engineering the exact algorithms neural networks learn. Most works in MI so far have studied behaviors and capabilities that are trivial and token-aligned. However, most capabilities are not that trivial, which advocates for the study of hidden representations inside these networks as the unit of analysis. We do a literature review, formalize representations for features and behaviors, highlight their importance and evaluation, and perform some basic exploration in the mechanistic interpretability …

ai models algorithms capabilities cs.ai cs.lg engineering frameworks hidden inside interpretability learn networks neural networks paper reverse-engineering study studying token