SAEs Are Good for Steering -- If You Select the Right Features

TL;DR

This paper demonstrates that selecting the right features in Sparse Autoencoders significantly improves model steering, achieving 2-3x better results by distinguishing input and output features with new scoring methods.

Contribution

The work introduces input and output scores to better identify features that influence model output, enhancing SAE-based steering methods.

Findings

Filtering features with high output scores improves steering effectiveness.

High input and output scores rarely co-occur in the same features.

SAEs become competitive with supervised methods after feature filtering.

Abstract

Sparse Autoencoders (SAEs) have been proposed as an unsupervised approach to learn a decomposition of a model's latent space. This enables useful applications such as steering - influencing the output of a model towards a desired concept - without requiring labeled data. Current methods identify SAE features to steer by analyzing the input tokens that activate them. However, recent work has highlighted that activations alone do not fully describe the effect of a feature on the model's output. In this work, we draw a distinction between two types of features: input features, which mainly capture patterns in the model's input, and output features, which have a human-understandable effect on the model's output. We propose input and output scores to characterize and locate these types of features, and show that high values for both scores rarely co-occur in the same features. These findings have practical implications: after filtering out features with low output scores, we obtain 2-3x improvements when steering with SAEs, making them competitive with supervised methods.