FADE: Why Bad Descriptions Happen to Good Features

TL;DR

FADE is a scalable, model-agnostic framework that automatically evaluates the alignment between features and descriptions in interpretability pipelines, highlighting challenges and guiding improvements.

Contribution

Introduces FADE, a novel evaluation framework with four metrics, to systematically assess feature-description alignment in interpretability of LLMs.

Findings

FADE reveals fundamental challenges in generating accurate feature descriptions.

SAEs pose more difficulties than MLP neurons in description alignment.

FADE's analysis provides insights into interpretability limitations and future directions.

Abstract

Recent advances in mechanistic interpretability have highlighted the potential of automating interpretability pipelines in analyzing the latent representations within LLMs. While this may enhance our understanding of internal mechanisms, the field lacks standardized evaluation methods for assessing the validity of discovered features. We attempt to bridge this gap by introducing FADE: Feature Alignment to Description Evaluation, a scalable model-agnostic framework for automatically evaluating feature-to-description alignment. FADE evaluates alignment across four key metrics - Clarity, Responsiveness, Purity, and Faithfulness - and systematically quantifies the causes of the misalignment between features and their descriptions. We apply FADE to analyze existing open-source feature descriptions and assess key components of automated interpretability pipelines, aiming to enhance the quality of descriptions. Our findings highlight fundamental challenges in generating feature descriptions, particularly for SAEs compared to MLP neurons, providing insights into the limitations and future directions of automated interpretability. We release FADE as an open-source package at: https://github.com/brunibrun/FADE