Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words
This work addresses the interpretability of large language models for researchers by proposing a new evaluation suite, though it is incremental as it builds on existing SAE methods without introducing a new paradigm.
The paper tackled the problem that traditional metrics for sparse autoencoders (SAEs) fail to assess their semantic representational power, particularly in distinguishing meanings of polysemous words, and found that optimizing for these metrics can confuse interpretability without improving monosemantic feature extraction.
Sparse autoencoders (SAEs) have gained a lot of attention as a promising tool to improve the interpretability of large language models (LLMs) by mapping the complex superposition of polysemantic neurons into monosemantic features and composing a sparse dictionary of words. However, traditional performance metrics like Mean Squared Error and L0 sparsity ignore the evaluation of the semantic representational power of SAEs -- whether they can acquire interpretable monosemantic features while preserving the semantic relationship of words. For instance, it is not obvious whether a learned sparse feature could distinguish different meanings in one word. In this paper, we propose a suite of evaluations for SAEs to analyze the quality of monosemantic features by focusing on polysemous words. Our findings reveal that SAEs developed to improve the MSE-L0 Pareto frontier may confuse interpretability, which does not necessarily enhance the extraction of monosemantic features. The analysis of SAEs with polysemous words can also figure out the internal mechanism of LLMs; deeper layers and the Attention module contribute to distinguishing polysemy in a word. Our semantics focused evaluation offers new insights into the polysemy and the existing SAE objective and contributes to the development of more practical SAEs.