Pith. sign in

REVIEW 10 cited by

MIB: A Mechanistic Interpretability Benchmark

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2504.13151 v2 pith:OOWKI5UA submitted 2025-04-17 cs.LG cs.AIcs.CL

MIB: A Mechanistic Interpretability Benchmark

classification cs.LG cs.AIcs.CL
keywords causalmethodslocalizationinterpretabilitymechanisticvariableattributionbenchmark
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

How can we know whether new mechanistic interpretability methods achieve real improvements? In pursuit of lasting evaluation standards, we propose MIB, a Mechanistic Interpretability Benchmark, with two tracks spanning four tasks and five models. MIB favors methods that precisely and concisely recover relevant causal pathways or causal variables in neural language models. The circuit localization track compares methods that locate the model components - and connections between them - most important for performing a task (e.g., attribution patching or information flow routes). The causal variable localization track compares methods that featurize a hidden vector, e.g., sparse autoencoders (SAEs) or distributed alignment search (DAS), and align those features to a task-relevant causal variable. Using MIB, we find that attribution and mask optimization methods perform best on circuit localization. For causal variable localization, we find that the supervised DAS method performs best, while SAE features are not better than neurons, i.e., non-featurized hidden vectors. These findings illustrate that MIB enables meaningful comparisons, and increases our confidence that there has been real progress in the field.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 10 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. WriteSAE: Sparse Autoencoders for Recurrent State

    cs.LG 2026-05 unverdicted novelty 8.0

    WriteSAE introduces sparse autoencoders with rank-1 matrix atoms for recurrent state updates, allowing replacement tests that outperform deletion on 92.4% of positions and a formula predicting logit changes with R²=0.98.

  2. WriteSAE: Sparse Autoencoders for Recurrent State

    cs.LG 2026-05 unverdicted novelty 8.0

    WriteSAE is the first sparse autoencoder that factors decoder atoms into the native d_k x d_v cache write shape of recurrent models and supplies a closed-form per-token logit shift for atom substitution.

  3. WriteSAE: Sparse Autoencoders for Recurrent State

    cs.LG 2026-05 unverdicted novelty 8.0

    WriteSAE decomposes recurrent model cache writes into substitutable atoms with a closed-form logit shift, achieving high substitution success and targeted behavioral installs on models like Qwen3.5 and Mamba-2.

  4. Conditional Co-Ablation: Recovering Self-Repair Backups in Transformer Circuits

    cs.LG 2026-07 unverdicted novelty 7.0

    Conditional Co-Ablation recovers self-repair backup heads in transformers by scoring conditional ablation growth, raising ROC-AUC from 0.33 to 0.91 on the IOI circuit and transferring to induction across models.

  5. From Circuit Evidence to Mechanistic Theory: An Inductive Logic Approach

    cs.LG 2026-05 unverdicted novelty 7.0

    Introduces Causal Functional Signatures grounded in causal evidence and ILP-learned architectural signatures to enable explicit, comparable, and portable mechanistic claims across model scales.

  6. WriteSAE: Sparse Autoencoders for Recurrent State

    cs.LG 2026-05 unverdicted novelty 7.0

    WriteSAE factors sparse autoencoder decoder atoms to the native d_k x d_v cache write shape in recurrent models, provides a closed-form logit shift, and demonstrates high success in atom substitution and behavioral ed...

  7. GKnow: Measuring the Entanglement of Gender Bias and Factual Gender

    cs.CL 2026-05 unverdicted novelty 7.0

    Gender bias and factual gender knowledge are severely entangled in language model circuits and neurons, making neuron ablation an unreliable method for debiasing.

  8. Temporal Preference Concepts and their Functions in a Large Language Model

    cs.LG 2026-05 conditional novelty 6.5

    Temporal preference in Qwen3-4B-Instruct-2507 localizes to layers 17–35 (especially L24 attention), has curved residual-stream geometry, is behaviorally unstable, and can be bidirectionally steered.

  9. ToxiREX: A Dataset on Toxic REasoning in ConteXt

    cs.CL 2026-06 unverdicted novelty 6.0

    ToxiREX is a new dataset of 128k Reddit comments in six languages with hierarchical annotations for implicit toxicity in conversational context based on an existing reasoning schema.

  10. Temporal Preference Concepts and their Functions in a Large Language Model

    cs.LG 2026-05 unverdicted novelty 5.0

    Causal localization via attribution and patching identifies a temporal preference subgraph in mid-to-upper layers of Qwen3-4B-Instruct-2507, with time-horizon geometry in the residual stream and initial evidence for s...