Evaluating Open-Source Sparse Autoencoders on Disentangling Factual Knowledge in GPT-2 Small
This work addresses the problem of evaluating interpretability methods for AI researchers, finding that current open-source SAEs are incremental and underperform existing approaches in disentangling knowledge.
The paper evaluated whether sparse autoencoders (SAEs) trained on GPT-2 small activations can disentangle factual knowledge about cities, such as separating country from continent information, using the RAVEL benchmark. Results showed that SAEs performed worse than neuron baselines and far below distributed alignment search (DAS) methods, indicating limited utility for causal analysis.
A popular new method in mechanistic interpretability is to train high-dimensional sparse autoencoders (SAEs) on neuron activations and use SAE features as the atomic units of analysis. However, the body of evidence on whether SAE feature spaces are useful for causal analysis is underdeveloped. In this work, we use the RAVEL benchmark to evaluate whether SAEs trained on hidden representations of GPT-2 small have sets of features that separately mediate knowledge of which country a city is in and which continent it is in. We evaluate four open-source SAEs for GPT-2 small against each other, with neurons serving as a baseline, and linear features learned via distributed alignment search (DAS) serving as a skyline. For each, we learn a binary mask to select features that will be patched to change the country of a city without changing the continent, or vice versa. Our results show that SAEs struggle to reach the neuron baseline, and none come close to the DAS skyline. We release code here: https://github.com/MaheepChaudhary/SAE-Ravel