AgentFlare

Scholar · mechanistic interpretability neural networks

Mechanistic Interpretability: Research Digest

10 foundational papers with 463 combined citations, plus the latest arXiv preprints.

Foundational papers10463 citations
Most citedA practical review of mechanistic int…244 citations · 2024
arXiv, Sep 202640 papers+8% vs Oct 2025
Newest preprint2026-10-26

AISA Scholar API · arXiv

arXiv submissions per monthPapers matching abs:"mechanistic interpretability" · last 12 complete months
monthly arXiv submissions: Oct 2025 37 → Sep 2026 40; high 77 (May 2026), low 23 (Dec 2025)19406081Oct 2025Mar 2026Sep 2026
402 · Pay-per-Crawl$0.02

You're reading the full report — free for humans and search engines. The same content ships as a structured JSON dataset (normalized facts, sources and citation arrays, no HTML parsing) so AI agents can consume it directly — metered per fetch via the open HTTP 402 payment protocol.

⚡ Agents: fetch .data.json · $0.02 → agents: crawlerAutoPrice=true · verify at https://esa.aisa.one/api/v1/access/verify

What’s new on arXiv

A12026-10-26cs.LG

Mechanistic Insights into LoRA: Layer Sparsity for Adaptive Fine-Tuning via Path Patching

Fei Zuo, Yizhou Huang, Kezhi Wang

Low-Rank Adaptation (LoRA) has emerged as the dominant paradigm for parameter-efficient fine-tuning of large language models. Standard all-layer LoRA applies adaptation throughout the transformer, despite heterogeneous contributions from its components. In this work, we conduct a systematic mechanistic interpretability analysis to understand how LoRA affects model internals. We observe that LoRA's adaptation effect is highly non-uniform, with significant component-wise and layer-wise sparsity. Most critically, we find that important layers are highly task-dependent, with limited overlap across tasks. Motivated by these insights, we propose Adaptive-LoRA, a calibration-driven method that automatically identifies task-specific layers via frozen-base zero-ablation, enabling efficient layer selection without exhaustive search. Extensive experiments on Qwen2.5-1.5B across 11 diverse reasoning tasks demonstrate that Adaptive-LoRA outperforms full-layer LoRA while using only 30% of layers, achieving a 3.5× reduction in adapter parameters. This work bridges mechanistic interpretability and practical efficiency, providing both fundamental insights and a principled approach for adaptive fine-tuning.

A22026-10-22cs.CL

The Geometry of Metaphor: Probing and Steering Metaphor Representations in Large Language Models

Shihui Li, Mette Andersen, Martha Lewis

Metaphor is a crucial aspect of human cognition. As such, we would like to understand how it is represented in large language models (LLMs) and how it might be controlled. We use mechanistic interpretability techniques to identify causal subspaces in a range of LLMs covering different architectures and sizes. We show that we are able to steer output on classification tasks across different parts of speech. We are also able to steer output on generation tasks, indicating that these techniques can be used to control the metaphoricity of LLM output.

A32026-10-20cs.CL

Language Impairment as a Lens onto Large Language Models

Muxuan He, Wang Bill Zhu, Yue Chen, Elsi Kaiser, Weizhi Du, Yao Du, Samuel Nastase, Yanjun Gao

Although large language models (LLMs) encode language in highly distributed spaces, they remain vulnerable to focal disruptions---a phenomenon strikingly similar to how localized brain damage causes clinical aphasia in humans. To investigate this theoretical tension, we ask whether LLMs possess a functional structure analogous to the human language system, by introducing a novel mechanistic interpretability framework inspired by aphasia. Through two complementary approaches, classifier-guided lesioning and continuous activation steering, we systematically induce aphasia-like breakdowns in LLMs. We find that language functions can be successfully isolated to discrete structural circuits and continuous representational vectors, yet these functions remain densely entangled with significant circuit overlap and vector similarities across distinct aphasic profiles. Furthermore, while steering generalizes robustly across downstream tasks, structural lesioning remains asymmetrically constrained. By bridging neuropsychology and mechanistic interpretability, we demonstrate that LLMs, functionally analogous to biological brains, rely on a densely entangled architecture that can be systematically mapped through clinically-informed interventions.

A42026-10-20cs.CL

Benchmarking Arabic Dialects for Math Problems

Noor Abo Mokh, Teresa Lynn, Nizar Habash, Ahmed Elshabrawy, Sara Alhajeri, Alaa Elsetohy, Preslav Nakov, Alham Fikri Aji

Mathematical reasoning remains underexplored for low-resource languages, including Dialectal Arabic. To bridge this gap, we introduce the DAR-MGSM (Dialectal Arabic Multilingual Grade School Math) Benchmark, the first manually curated benchmark for evaluating mathematical reasoning in Dialectal Arabic. DAR-MGSM consists of math word problems (MWPs) requiring multi-step reasoning and covers five major Arabic dialects alongside Modern Standard Arabic (MSA). The benchmark comprises two parallel versions: translated-only and translated & culturally localized. The latter version adapts entities and references to regional varieties. We evaluate eleven Arabic-centric and multilingual large language models on DAR-MGSM. The results show that the performance generally declines on the localized version, highlighting the challenges introduced by culturally grounded adaptations. However, the magnitude of this effect varies across models, with Arabic-centric models exhibiting different levels of sensitivity to localization. A mechanistic interpretability analysis shows that translated and localized variants of the same MWPs often induce distinct internal reasoning trajectories, even when aggregate accuracy remains similar. Overall, our results show that mathematical reasoning performance depends on both linguistic variety and cultural context.

A52026-10-20cs.CL

Are Linguistic and Arithmetic Circuits Dissociated in LLMs?

Riku Kisako, Tatsuki Kuribayashi, Ryohei Sasano

Large language models (LLMs) exhibit both knowledge of language, rendered by, for example, grammatical judgment ability, and abilities to perform tasks through language, such as math-word problem solving. Motivated by neurolinguistic studies of their neural dissociation in human brains, in this study, we investigate the overlap of their circuits in large language models (LLMs). Specifically, we examine the overlap and causal effects between the circuits extracted for linguistic and arithmetic tasks, using controlled materials and circuit discovery methods from the mechanistic interpretability field. Our experiments with five LLMs show that circuit similarity is mostly explained by task domains (linguistic vs. arithmetic), rather than by their input and output vocabularies or by the syntactic similarities between natural-language sentences and arithmetic expressions. This finding suggests that LLMs may process linguistic and arithmetic information using largely separate neural circuits, highlighting an ability-based modular organization of circuits within these models.

A62026-10-19cs.AI

Correcting Attribution Patching with a Few Exact Measurements

Luyang Zhang, Minxing Zheng, Jialu Wang

Mechanistic interpretability asks which language model components matter for a behavior, and activation patching answers by changing one component's activation and measuring its effect, at a cost growing with components and prompts. Attribution patching estimates these effects from one gradient pass, but its estimates can be far off, and verifying each one restores the cost. Can the structure of these errors reduce this cost? We propose Guided-AP, which uses a few exact measurements to reveal attribution patching's errors and correct unmeasured estimates. On a single prompt, many components' errors share one sign, so a small exact sample yields one correction for the rest. Over many prompts of one task, components are misestimated similarly, so a few measured prompts show what to verify and how to correct each component's average effect. Our analysis shows that choosing what to verify needs error size, while correction also needs sign. On a single prompt, the correction lowers error by up to 26% across three behaviors at the compute of verifying the largest estimates and also on larger models. Over many prompts, learned corrections cut the error of average effects on factual completion by 75% to 79%, repaying their cost in long analyses.

A72026-10-18cs.CC

Formally Verified Autoresearch for Analyzing the Expressivity of Multi-Head Attention

Karthik Viswanathan, Plamen Dimitrov, Stefano Rocca, Sarayu Sirikonda

Multi-head attention is a central component of the transformer architecture, yet how its expressivity depends on the number of heads remains poorly understood. We study a one-layer, attention-only transformer with additive positional embeddings and define the head complexity H^ast(f) of a Boolean function f as the minimum number of attention heads required to represent it. We derive an exact linear-fractional normal form for head complexity, prove a general threshold-degree lower bound, and give an explicit ten-bit function for which this bound is strict. For symmetric Boolean functions, we independently show that head complexity equals the number of times the function changes value between consecutive Hamming-weight levels, even when positional embeddings are allowed. This agrees with the concurrent characterization of Hsu and Xu (arXiv:2608.04243) in a model without positional embeddings. The symmetric lower bound combines our general threshold-degree bound with the classical result of Aspnes et al. (1994). These results were discovered and formalized by a frontier model through conjecture generation, theorem proving, and machine verification, while human researchers asked the high-level questions and reviewed the generated theorems. We make the Lean formalization and research traces publicly available to support agent-assisted theorem proving and formal verification in mechanistic interpretability. This work provides a proof of concept for using the capabilities of frontier models, grounded by formal verification, to develop an understanding of transformer architectures.

A82026-10-15cs.CV

Spurious Attention Heads in Vision Transformers: Mechanistic Identification of Annotation-Driven Bias in Medical Chest X-Ray Analysis

Joshua Muhumuza, Cornelius Kalenzi

Vision Transformers deployed in medical imaging exploit spurious correlations, yet the internal mechanisms remain unidentified and impossible to fix. We apply mechanistic interpretability to locate one such mechanism in ViT-Base 16: a single attention head (L8H10) that functions as a dedicated annotation-reader on chest X-rays, attending to corner text rather than lung pathology in 62% of 200 COVID-19 images. Causal validation via inpainting confirms annotation content rather than luminance or edge density drives the behaviour, increasing lung attention 1.52 times upon removal in affected images (p<0.0001). The finding replicates across NIH ChestX-ray14 (79%), DeiT-Base (88%), and ViT-Large 16 (80%), establishing spurious annotation-attending heads as a consistent property of ViT-family architectures across two datasets and three architectures. Full fine-tuning does not suppress the behaviour and may entrench it: 98.4% of spurious images remain spurious and the spurious rate rises marginally, as the shortcut is reinforced rather than removed. Critically, single-head ablation produces p=0.340 and multi-head ablation of all three identified carriers produces p=0.532, meaning a causally verified annotation-reading circuit is completely invisible to output-level evaluation. Mechanistic auditing finds what behavioural testing cannot.

Foundational papers

  1. 2004
    Knowledge-driven interpretable neural networks provide mechanistic insight0 citations
  2. 2022
    Tutorial on Mechanistic Interpretability0 citations
  3. 2024
    A practical review of mechanistic interpretability for transformer-based language models244 citations
  4. 2024
    Mechanistic?119 citations
  5. 2025
    Everything, everywhere, all at once: Is mechanistic interpretability identifiable?45 citations
  6. 2026
    A Unifying Perspective on Language Model Representations: From Filler-Role Structure to Mechanistic Interpretability1 citations
  7. 2026-10
    Latest on arXiv: Mechanistic Insights into LoRA: Layer Sparsity for Adaptive Fine-Tuning via Path Patchingcs.LG
Citations per paperCitation counts via the AISA Scholar API
citations: R1 Bridging the black bo… 49, R2 Everything, everywher… 45, R3 A practical review of… 244, R4 Unboxing the Black Bo… 1, R5 Training neural netwo… 3, R6 Mechanistic? 119, R7 Tutorial on Mechanist… 0, R8 Mechanistic Interpret… 1, R9 A Unifying Perspectiv… 1, R10 Knowledge-driven inte… 0R1 Bridging the black bo…49R2 Everything, everywher…45R3 A practical review of…244R4 Unboxing the Black Bo…1R5 Training neural netwo…3R6 Mechanistic?119R7 Tutorial on Mechanist…0R8 Mechanistic Interpret…1R9 A Unifying Perspectiv…1R10 Knowledge-driven inte…0
R62024119 citations

Mechanistic?

N Saphra, S Wiegreffe

… To understand this semantic drift, we present a history of the NLP interpretability community and the formation of the separate, parallel mechanistic interpretability community. Finally, we …

R720220 citations

Tutorial on Mechanistic Interpretability

C Chen, M Heuss, C Eickhoff

… It can be narrowly defined as “[the discovery of] causal mechanisms explaining all or some part of the change from neural network input to output at the level of intermediate model …

How this page is made. Foundational papers and citation counts come from academic search via the AISA Scholar API, with abstracts read from each paper’s page; recent preprints and monthly volumes come from the public arXiv API (metadata CC0). The digest is written by a language model from those abstracts only — every theme links the papers it draws on, and figures not in the inputs are removed.

Sources & citations

  1. Bridging the black box: a survey on mechanistic interpretability in AI 49 citations
  2. Everything, everywhere, all at once: Is mechanistic interpretability identifiable? 45 citations
  3. A practical review of mechanistic interpretability for transformer-based language models 244 citations
  4. Unboxing the Black Box: A Survey on Mechanistic Interpretability for Algorithmic Understanding of Neural Networks 1 citations
  5. Training neural networks on mechanistic simulations improves scientific inference 3 citations
  6. Mechanistic? 119 citations
  7. Tutorial on Mechanistic Interpretability 0 citations
  8. Mechanistic Interpretability of ReLU Neural Networks Through Piecewise-Affine Mapping 1 citations
  9. A Unifying Perspective on Language Model Representations: From Filler-Role Structure to Mechanistic Interpretability 1 citations
  10. Knowledge-driven interpretable neural networks provide mechanistic insight 0 citations
  11. Mechanistic Insights into LoRA: Layer Sparsity for Adaptive Fine-Tuning via Path Patching arXiv · 2026-10-26
  12. The Geometry of Metaphor: Probing and Steering Metaphor Representations in Large Language Models arXiv · 2026-10-22
  13. Language Impairment as a Lens onto Large Language Models arXiv · 2026-10-20
  14. Benchmarking Arabic Dialects for Math Problems arXiv · 2026-10-20
  15. Are Linguistic and Arithmetic Circuits Dissociated in LLMs? arXiv · 2026-10-20
  16. Correcting Attribution Patching with a Few Exact Measurements arXiv · 2026-10-19
  17. Formally Verified Autoresearch for Analyzing the Expressivity of Multi-Head Attention arXiv · 2026-10-18
  18. Spurious Attention Heads in Vision Transformers: Mechanistic Identification of Annotation-Driven Bias in Medical Chest X-Ray Analysis arXiv · 2026-10-15

Related