Enhancing Automated Interpretability with Output-Centric Feature Descriptions

TL;DR

This paper introduces output-centric methods for generating feature descriptions in large language models, improving causal interpretability and enabling the discovery of previously inactive features.

Contribution

It proposes novel output-centric techniques for feature description generation that better capture causal effects on outputs, enhancing interpretability of LLM features.

Findings

Output-centric descriptions outperform input-centric ones in causal evaluations.

Combining input and output descriptions yields the best interpretability results.

Output-centric methods can identify inputs activating previously 'dead' features.

Abstract

Automated interpretability pipelines generate natural language descriptions for the concepts represented by features in large language models (LLMs), such as plants or the first word in a sentence. These descriptions are derived using inputs that activate the feature, which may be a dimension or a direction in the model's representation space. However, identifying activating inputs is costly, and the mechanistic role of a feature in model behavior is determined both by how inputs cause a feature to activate and by how feature activation affects outputs. Using steering evaluations, we reveal that current pipelines provide descriptions that fail to capture the causal effect of the feature on outputs. To fix this, we propose efficient, output-centric methods for automatically generating feature descriptions. These methods use the tokens weighted higher after feature stimulation or the highest weight tokens after applying the vocabulary "unembedding" head directly to the feature. Our output-centric descriptions better capture the causal effect of a feature on model outputs than input-centric descriptions, but combining the two leads to the best performance on both input and output evaluations. Lastly, we show that output-centric descriptions can be used to find inputs that activate features previously thought to be "dead".