nucleation

LIT-1532: Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models

1 min read

Properties5
StatusActive — read 2026-10-26 (NOTE-1187) in full, from the arXiv copy (v3, the ICLR 2025 version); worth reading as the method that turned sparse-autoencoder features into circuits, by linear approximations of indirect effects over features and autoencoder errors together, and for its own evidence that the features alone are not the circuit: with the error nodes removed, its feature circuits recover little of the behaviour (THEORY-1007). Read it knowing that faithfulness is scored on the latter two thirds of each circuit only, that the interpretability of the circuits is the authors' annotation, that the half-performance figure for Gemma neurons lies beyond its plotted range, and that its debiasing method is ML practice, which this record does not hold.
Version2
Tagsrepresentation-learning, causality, anthology-candidate
Cited as “paper” byNOTE-1187 — Marks, Rager, Michaud, Belinkov, Bau and Mueller — Sparse feature circuits
Cited as “source” byTHEORY-1007 — On four subject–verb agreement tasks in Pythia-70M and Gemma-2-2B, circuits of sparse-autoencoder features and autoencoder error nodes, selected by linear estimates of indirect effect, carried half or more of the behaviour on held-out pairs with about 100 and 500 nodes, where circuits of neurons needed about 1,500 and more than 30,000; with the error nodes removed, the same feature circuits carried little of it, under about a quarter in Pythia and almost none in Gemma, THEORY-1014 — In a language model, a behaviour that looks like one mechanism, whether picked out by one contrast or by one cluster of next-token predictions, is often carried by several distinct pathways of sparse features that converge on the output without fully intersecting, as in Pythia-70M's agreement across a relative clause, its prediction of "to" before an infinitival object, and its incrementing of sequences

Graph View

Marks, S., Rager, C., Michaud, E. J., Belinkov, Y., Bau, D., and Mueller, A. (2025). Sparse feature circuits: Discovering and editing interpretable causal graphs in language models. ICLR 2025. arXiv:2403.19647.

Read in full on 2026-10-26 (NOTE-1187). Two accounts were drawn from it: THEORY-1007 and THEORY-1014.

Why this is here

Registered on 2026-10-26 at the owner’s request, when the record was found to hold almost none of the core circuits literature. It is the bridge from features to circuits: the record held sparse-autoencoder features as units (LIT-1545, LIT-1533) and circuits over heads (LIT-1538), but not a circuit built of features. Its indirect effects with the integrated-gradient approximation are what the record’s crosscoder reading builds on (LIT-1555, THEORY-1004), and its last author leads MIB (LIT-1550). It bears on CLAIM-454, THEORY-968 and QUESTION-275.

Backlinks