Language Model Circuits Are Sparse in the Neuron Basis

Aryaman Arora, Zhengxuan Wu, Jacob Steinhardt, Sarah Schwettmann
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:3712-3753, 2026.

Abstract

The high-level concepts that a neural network uses to perform computation need not be aligned to individual neurons (Smolensky, 1986). Language model interpretability research has thus turned to techniques which decompose the neuron basis into more interpretable units of model computation, such as sparse autoencoders (SAEs). However, not all neuron-based representations are uninterpretable. For the first time, we empirically show that MLP neurons are as sparse a feature basis as SAEs. We use this finding to develop an end-to-end gradient-based attribution pipeline for circuit tracing on the MLP neuron basis, which surfaces causally effective neurons on a variety of tasks. On a standard subject-verb agreement benchmark (Marks et al., 2025), a circuit of $\approx 10^2$ MLP neurons is enough to control model behaviour. On the multi-hop city-state-capital task from Lindsey et al. (2025), we find a circuit in which small sets of neurons encode specific latent reasoning steps (e.g. mapping a city to its state), and can be steered to change the model’s output. This work thus advances automated interpretability of language models without imposing additional training costs.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-arora26c, title = {Language Model Circuits Are Sparse in the Neuron Basis}, author = {Arora, Aryaman and Wu, Zhengxuan and Steinhardt, Jacob and Schwettmann, Sarah}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {3712--3753}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/arora26c/arora26c.pdf}, url = {https://proceedings.mlr.press/v306/arora26c.html}, abstract = {The high-level concepts that a neural network uses to perform computation need not be aligned to individual neurons (Smolensky, 1986). Language model interpretability research has thus turned to techniques which decompose the neuron basis into more interpretable units of model computation, such as sparse autoencoders (SAEs). However, not all neuron-based representations are uninterpretable. For the first time, we empirically show that MLP neurons are as sparse a feature basis as SAEs. We use this finding to develop an end-to-end gradient-based attribution pipeline for circuit tracing on the MLP neuron basis, which surfaces causally effective neurons on a variety of tasks. On a standard subject-verb agreement benchmark (Marks et al., 2025), a circuit of $\approx 10^2$ MLP neurons is enough to control model behaviour. On the multi-hop city-state-capital task from Lindsey et al. (2025), we find a circuit in which small sets of neurons encode specific latent reasoning steps (e.g. mapping a city to its state), and can be steered to change the model’s output. This work thus advances automated interpretability of language models without imposing additional training costs.} }
Endnote
%0 Conference Paper %T Language Model Circuits Are Sparse in the Neuron Basis %A Aryaman Arora %A Zhengxuan Wu %A Jacob Steinhardt %A Sarah Schwettmann %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-arora26c %I PMLR %P 3712--3753 %U https://proceedings.mlr.press/v306/arora26c.html %V 306 %X The high-level concepts that a neural network uses to perform computation need not be aligned to individual neurons (Smolensky, 1986). Language model interpretability research has thus turned to techniques which decompose the neuron basis into more interpretable units of model computation, such as sparse autoencoders (SAEs). However, not all neuron-based representations are uninterpretable. For the first time, we empirically show that MLP neurons are as sparse a feature basis as SAEs. We use this finding to develop an end-to-end gradient-based attribution pipeline for circuit tracing on the MLP neuron basis, which surfaces causally effective neurons on a variety of tasks. On a standard subject-verb agreement benchmark (Marks et al., 2025), a circuit of $\approx 10^2$ MLP neurons is enough to control model behaviour. On the multi-hop city-state-capital task from Lindsey et al. (2025), we find a circuit in which small sets of neurons encode specific latent reasoning steps (e.g. mapping a city to its state), and can be steered to change the model’s output. This work thus advances automated interpretability of language models without imposing additional training costs.
APA
Arora, A., Wu, Z., Steinhardt, J. & Schwettmann, S.. (2026). Language Model Circuits Are Sparse in the Neuron Basis. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:3712-3753 Available from https://proceedings.mlr.press/v306/arora26c.html.

Related Material