dorsal/arxiv
View SchemaDo Sparse Autoencoders Identify Reasoning Features in Language Models?
| Authors | George Ma, Zhongyuan Liang, Irene Y. Chen, Somayeh Sojoudi |
|---|---|
| Categories | |
| ArXiv ID | 2601.05679vv5 |
| URL | https://arxiv.org/abs/2601.05679 |
| License | http://creativecommons.org/licenses/by/4.0/ |
Abstract
We investigate whether sparse autoencoders (SAEs) identify genuine reasoning features in large language models (LLMs). We first show through a simple theoretical analysis that $\ell_1$-regularized SAEs are intrinsically biased toward low-dimensional patterns, providing a mechanistic explanation for why shallow linguistic cues may be preferentially captured over distributed reasoning behaviors. Motivated by this bias, we introduce a falsification-oriented evaluation framework that combines causal token injection and LLM-guided falsification to test whether feature activation reflects reasoning processes or superficial linguistic correlates. Across 20 configurations spanning multiple model families, layers, and reasoning datasets, we find that features identified by contrastive methods are highly sensitive to token-level interventions, with 45% to 90% activating when a small number of associated tokens are injected into non-reasoning text. For the remaining features, LLM-guided falsification consistently produces non-reasoning inputs that activate the feature and reasoning inputs that do not, with no analyzed feature satisfying our criteria for genuine reasoning behavior. Steering these features yields no improvements in benchmark performance. Overall, our results suggest that SAE features identified by current contrastive approaches primarily capture linguistic correlates of reasoning rather than the underlying reasoning computations themselves. Code is available at https://github.com/GeorgeMLP/reasoning-probing.
{
"annotation_id": "05d6e996-ba0b-4c87-8a7c-caed1bb69801",
"date_created": "2026-02-17T05:53:04.266000Z",
"date_modified": "2026-02-17T05:53:04.266000Z",
"file_hash": "878b343f897ece28513ed1180fc87f5e12d0734831badf405848bbd8c3e2a466",
"private": false,
"record": {
"abstract": "We investigate whether sparse autoencoders (SAEs) identify genuine reasoning features in large language models (LLMs). We first show through a simple theoretical analysis that $\\ell_1$-regularized SAEs are intrinsically biased toward low-dimensional patterns, providing a mechanistic explanation for why shallow linguistic cues may be preferentially captured over distributed reasoning behaviors. Motivated by this bias, we introduce a falsification-oriented evaluation framework that combines causal token injection and LLM-guided falsification to test whether feature activation reflects reasoning processes or superficial linguistic correlates. Across 20 configurations spanning multiple model families, layers, and reasoning datasets, we find that features identified by contrastive methods are highly sensitive to token-level interventions, with 45% to 90% activating when a small number of associated tokens are injected into non-reasoning text. For the remaining features, LLM-guided falsification consistently produces non-reasoning inputs that activate the feature and reasoning inputs that do not, with no analyzed feature satisfying our criteria for genuine reasoning behavior. Steering these features yields no improvements in benchmark performance. Overall, our results suggest that SAE features identified by current contrastive approaches primarily capture linguistic correlates of reasoning rather than the underlying reasoning computations themselves. Code is available at https://github.com/GeorgeMLP/reasoning-probing.",
"arxiv_id": "2601.05679",
"authors": [
"George Ma",
"Zhongyuan Liang",
"Irene Y. Chen",
"Somayeh Sojoudi"
],
"categories": [
"cs.LG"
],
"license": "http://creativecommons.org/licenses/by/4.0/",
"title": "Do Sparse Autoencoders Identify Reasoning Features in Language Models?",
"url": "https://arxiv.org/abs/2601.05679",
"version": "v5"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "6cd1dace-690f-4b63-ad27-a678f7ed42be",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}