Large Language Models (LLMs) are traditionally viewed as black-box algorithms, therefore reducing trustworthiness and obscuring potential approaches to increasing performance on downstream tasks. In this work, we apply an effective LLM decomposition method using a dictionary-learning approach with sparse autoencoders. This helps extract monosemantic features from polysemantic LLM neurons. Remarkably, our work identifies model-internal misunderstanding, allowing the automatic reformulation of the prompts with additional annotations to improve the interpretation by LLMs. Moreover, this approach demonstrates a significant performance improvement in downstream tasks, such as mathematical reasoning and metaphor detection.
@article{arxiv.2507.06427,
title = {Exploring Task Performance with Interpretable Models via Sparse Auto-Encoders},
author = {Shun Wang and Tyler Loakman and Youbo Lei and Yi Liu and Bohao Yang and Yuting Zhao and Dong Yang and Chenghua Lin},
journal= {arXiv preprint arXiv:2507.06427},
year = {2025}
}