new

Get trending papers in your email inbox!

Subscribe

Daily Papers

byAK and the research community

Nov 19

Using Mechanistic Interpretability to Craft Adversarial Attacks against Large Language Models

Traditional white-box methods for creating adversarial perturbations against LLMs typically rely only on gradient computation from the targeted model, ignoring the internal mechanisms responsible for attack success or failure. Conversely, interpretability studies that analyze these internal mechanisms lack practical applications beyond runtime interventions. We bridge this gap by introducing a novel white-box approach that leverages mechanistic interpretability techniques to craft practical adversarial inputs. Specifically, we first identify acceptance subspaces - sets of feature vectors that do not trigger the model's refusal mechanisms - then use gradient-based optimization to reroute embeddings from refusal subspaces to acceptance subspaces, effectively achieving jailbreaks. This targeted approach significantly reduces computation cost, achieving attack success rates of 80-95\% on state-of-the-art models including Gemma2, Llama3.2, and Qwen2.5 within minutes or even seconds, compared to existing techniques that often fail or require hours of computation. We believe this approach opens a new direction for both attack research and defense development. Furthermore, it showcases a practical application of mechanistic interpretability where other methods are less efficient, which highlights its utility. The code and generated datasets are available at https://github.com/Sckathach/subspace-rerouting.

  • 3 authors
·
Mar 8, 2025 2

Exemplar Partitioning for Mechanistic Interpretability

We introduce Exemplar Partitioning (EP), an unsupervised method for constructing interpretable feature dictionaries from large language model activations with sim 10^3times fewer tokens than comparable sparse autoencoders (SAEs). An EP dictionary is a Voronoi partition of activation space, built by leader-clustering streamed activations within a distance threshold. Each region is anchored by an observed exemplar that serves as both its membership criterion and intervention direction; dictionary size is not prespecified, but determined by the activation geometry at that threshold. Because exemplars are observed rather than learned, dictionaries built from the same data stream are directly comparable across layers, models, and training checkpoints. We characterise EP as an interpretability object via targeted demonstrations of properties newly accessible through this construction, plus one head-to-head benchmark. In Gemma-2-2B, EP dictionary regions are interpretable and support causal interventions: refusal in instruction-tuned Gemma concentrates in a region whose exemplar ablation can collapse held-out refusal. Cross-checkpoint matching between base and instruction-tuned dictionaries separates the directions preserved through finetuning from those introduced by it. EP regions and Gemma Scope SAE features decompose activation space differently but agree on a shared core: sim20% of EP regions match an SAE feature at F_1 > 0.5, and EP one-hot probes retain sim97% of raw-activation probe accuracy at ell_0 = 1. Nearest-exemplar distance provides a free out-of-distribution signal at inference. On AxBench latent concept detection at Gemma-2-2B-it L20, EP at p_1 reaches mean AUROC 0.881, +0.126 over the canonical GemmaScope SAE leaderboard entry and within 0.030 of SAE-A's 0.911, at sim 10^3times less build compute.

  • 1 authors
·
May 14

Counting Circuits: Mechanistic Interpretability of Visual Reasoning in Large Vision-Language Models

Counting serves as a simple but powerful test of a Large Vision-Language Model's (LVLM's) reasoning; it forces the model to identify each individual object and then add them all up. In this study, we investigate how LVLMs implement counting using controlled synthetic and real-world benchmarks, combined with mechanistic analyses. Our results show that LVLMs display a human-like counting behavior, with precise performance on small numerosities and noisy estimation for larger quantities. We introduce two novel interpretability methods, Visual Activation Patching and HeadLens, and use them to uncover a structured "counting circuit" that is largely shared across a variety of visual reasoning tasks. Building on these insights, we propose a lightweight intervention strategy that exploits simple and abundantly available synthetic images to fine-tune arbitrary pretrained LVLMs exclusively on counting. Despite the narrow scope of this fine-tuning, the intervention not only enhances counting accuracy on in-distribution synthetic data, but also yields an average improvement of +8.36% on out-of-distribution counting benchmarks and an average gain of +1.54% on complex, general visual reasoning tasks for Qwen2.5-VL. These findings highlight the central, influential role of counting in visual reasoning and suggest a potential pathway for improving overall visual reasoning capabilities through targeted enhancement of counting mechanisms.

  • 10 authors
·
Mar 18

Disentangling Recall and Reasoning in Transformer Models through Layer-wise Attention and Activation Analysis

Transformer-based language models excel at both recall (retrieving memorized facts) and reasoning (performing multi-step inference), but whether these abilities rely on distinct internal mechanisms remains unclear. Distinguishing recall from reasoning is crucial for predicting model generalization, designing targeted evaluations, and building safer interventions that affect one ability without disrupting the other.We approach this question through mechanistic interpretability, using controlled datasets of synthetic linguistic puzzles to probe transformer models at the layer, head, and neuron level. Our pipeline combines activation patching and structured ablations to causally measure component contributions to each task type. Across two model families (Qwen and LLaMA), we find that interventions on distinct layers and attention heads lead to selective impairments: disabling identified "recall circuits" reduces fact-retrieval accuracy by up to 15\% while leaving reasoning intact, whereas disabling "reasoning circuits" reduces multi-step inference by a comparable margin. At the neuron level, we observe task-specific firing patterns, though these effects are less robust, consistent with neuronal polysemanticity.Our results provide the first causal evidence that recall and reasoning rely on separable but interacting circuits in transformer models. These findings advance mechanistic interpretability by linking circuit-level structure to functional specialization and demonstrate how controlled datasets and causal interventions can yield mechanistic insights into model cognition, informing safer deployment of large language models.

  • 6 authors
·
Oct 3, 2025

The Local Interaction Basis: Identifying Computationally-Relevant and Sparsely Interacting Features in Neural Networks

Mechanistic interpretability aims to understand the behavior of neural networks by reverse-engineering their internal computations. However, current methods struggle to find clear interpretations of neural network activations because a decomposition of activations into computational features is missing. Individual neurons or model components do not cleanly correspond to distinct features or functions. We present a novel interpretability method that aims to overcome this limitation by transforming the activations of the network into a new basis - the Local Interaction Basis (LIB). LIB aims to identify computational features by removing irrelevant activations and interactions. Our method drops irrelevant activation directions and aligns the basis with the singular vectors of the Jacobian matrix between adjacent layers. It also scales features based on their importance for downstream computation, producing an interaction graph that shows all computationally-relevant features and interactions in a model. We evaluate the effectiveness of LIB on modular addition and CIFAR-10 models, finding that it identifies more computationally-relevant features that interact more sparsely, compared to principal component analysis. However, LIB does not yield substantial improvements in interpretability or interaction sparsity when applied to language models. We conclude that LIB is a promising theory-driven approach for analyzing neural networks, but in its current form is not applicable to large language models.

  • 10 authors
·
May 17, 2024

Automated Circuit Interpretation via Probe Prompting

Mechanistic interpretability aims to understand neural networks by identifying which learned features mediate specific behaviors. Attribution graphs reveal these feature pathways, but interpreting them requires extensive manual analysis -- a single prompt can take approximately 2 hours for an experienced circuit tracer. We present probe prompting, an automated pipeline that transforms attribution graphs into compact, interpretable subgraphs built from concept-aligned supernodes. Starting from a seed prompt and target logit, we select high-influence features, generate concept-targeted yet context-varying probes, and group features by cross-prompt activation signatures into Semantic, Relationship, and Say-X categories using transparent decision rules. Across five prompts including classic "capitals" circuits, probe-prompted subgraphs preserve high explanatory coverage while compressing complexity (Completeness 0.83, mean across circuits; Replacement 0.54). Compared to geometric clustering baselines, concept-aligned groups exhibit higher behavioral coherence: 2.3x higher peak-token consistency (0.425 vs 0.183) and 5.8x higher activation-pattern similarity (0.762 vs 0.130), despite lower geometric compactness. Entity-swap tests reveal a layerwise hierarchy: early-layer features transfer robustly (64% transfer rate, mean layer 6.3), while late-layer Say-X features specialize for output promotion (mean layer 16.4), supporting a backbone-and-specialization view of transformer computation. We release code (https://github.com/peppinob-ol/attribution-graph-probing), an interactive demo (https://huggingface.co/spaces/Peppinob/attribution-graph-probing), and minimal artifacts enabling immediate reproduction and community adoption.

  • 1 authors
·
Nov 10, 2025

Pando: Do Interpretability Methods Work When Models Won't Explain Themselves?

Mechanistic interpretability is often motivated for alignment auditing, where a model's verbal explanations can be absent, incomplete, or misleading. Yet many evaluations do not control whether black-box prompting alone can recover the target behavior, so apparent gains from white-box tools may reflect elicitation rather than internal signal; we call this the elicitation confounder. We introduce Pando, a model-organism benchmark that breaks this confound via an explanation axis: models are trained to produce either faithful explanations of the true rule, no explanation, or confident but unfaithful explanations of a disjoint distractor rule. Across 720 finetuned models implementing hidden decision-tree rules, agents predict held-out model decisions from 10 labeled query-response pairs, optionally augmented with one interpretability tool output. When explanations are faithful, black-box elicitation matches or exceeds all white-box methods; when explanations are absent or misleading, gradient-based attribution improves accuracy by 3-5 percentage points, and relevance patching, RelP, gives the largest gains, while logit lens, sparse autoencoders, and circuit tracing provide no reliable benefit. Variance decomposition suggests gradients track decision computation, which fields causally drive the output, whereas other readouts are dominated by task representation, biases toward field identity and value. We release all models, code, and evaluation infrastructure.

  • 5 authors
·
Apr 12

Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patching

Mechanistic interpretability aims to understand model behaviors in terms of specific, interpretable features, often hypothesized to manifest as low-dimensional subspaces of activations. Specifically, recent studies have explored subspace interventions (such as activation patching) as a way to simultaneously manipulate model behavior and attribute the features behind it to given subspaces. In this work, we demonstrate that these two aims diverge, potentially leading to an illusory sense of interpretability. Counterintuitively, even if a subspace intervention makes the model's output behave as if the value of a feature was changed, this effect may be achieved by activating a dormant parallel pathway leveraging another subspace that is causally disconnected from model outputs. We demonstrate this phenomenon in a distilled mathematical example, in two real-world domains (the indirect object identification task and factual recall), and present evidence for its prevalence in practice. In the context of factual recall, we further show a link to rank-1 fact editing, providing a mechanistic explanation for previous work observing an inconsistency between fact editing performance and fact localization. However, this does not imply that activation patching of subspaces is intrinsically unfit for interpretability. To contextualize our findings, we also show what a success case looks like in a task (indirect object identification) where prior manual circuit analysis informs an understanding of the location of a feature. We explore the additional evidence needed to argue that a patched subspace is faithful.

  • 3 authors
·
Nov 28, 2023

Using Degeneracy in the Loss Landscape for Mechanistic Interpretability

Mechanistic Interpretability aims to reverse engineer the algorithms implemented by neural networks by studying their weights and activations. An obstacle to reverse engineering neural networks is that many of the parameters inside a network are not involved in the computation being implemented by the network. These degenerate parameters may obfuscate internal structure. Singular learning theory teaches us that neural network parameterizations are biased towards being more degenerate, and parameterizations with more degeneracy are likely to generalize further. We identify 3 ways that network parameters can be degenerate: linear dependence between activations in a layer; linear dependence between gradients passed back to a layer; ReLUs which fire on the same subset of datapoints. We also present a heuristic argument that modular networks are likely to be more degenerate, and we develop a metric for identifying modules in a network that is based on this argument. We propose that if we can represent a neural network in a way that is invariant to reparameterizations that exploit the degeneracies, then this representation is likely to be more interpretable, and we provide some evidence that such a representation is likely to have sparser interactions. We introduce the Interaction Basis, a tractable technique to obtain a representation that is invariant to degeneracies from linear dependence of activations or Jacobians.

  • 8 authors
·
May 17, 2024

Mechanist: AI as a Scientific Instrument for Discovering the Mechanisms of Intelligence

AI models have achieved remarkable success across diverse domains, yet the mechanisms underlying their capabilities and the risks they may pose remain poorly understood. As AI development becomes faster and increasingly automated, mechanistic exploration remains largely manual, widening the gap between what models can do and our ability to understand and control them. To bridge this gap, we introduce Mechanist, an agentic system that uses AI as a scientific instrument for the autonomous discovery of mechanisms underlying AI intelligence. To support autonomous mechanistic discovery, we construct an interpretability-focused knowledge graph of approximately 13,000 papers and integrate it with a multidisciplinary database of 43 million papers spanning 26 fields. We further curate a library of 32 foundational methods for mechanism analysis, causal intervention, and validation. Compared with Claude Code and existing AI-scientist systems, Mechanist generates more valuable mechanism hypotheses and executes experiments more reliably. Mechanist also demonstrates a progression from discovering model behaviors to explaining and controlling AI models. Specifically, Mechanist first uncovers a counterintuitive safety risk in scientific laboratories, showing that unsafe traits can transfer across modalities through apparently safe training data. Mechanist then develops a mechanism theory of belief, revealing how models represent world knowledge, form beliefs, infer the beliefs of others, and how these mechanisms emerge during pretraining. Finally, Mechanist translates these mechanistic insights into practical interventions that improve model performance across diverse scenarios and steer scientific foundation models toward generating DNA sequences with specified properties.

zjunlp ZJUNLP
·
Aug 11 2

Decomposing MLP Activations into Interpretable Features via Semi-Nonnegative Matrix Factorization

A central goal for mechanistic interpretability has been to identify the right units of analysis in large language models (LLMs) that causally explain their outputs. While early work focused on individual neurons, evidence that neurons often encode multiple concepts has motivated a shift toward analyzing directions in activation space. A key question is how to find directions that capture interpretable features in an unsupervised manner. Current methods rely on dictionary learning with sparse autoencoders (SAEs), commonly trained over residual stream activations to learn directions from scratch. However, SAEs often struggle in causal evaluations and lack intrinsic interpretability, as their learning is not explicitly tied to the computations of the model. Here, we tackle these limitations by directly decomposing MLP activations with semi-nonnegative matrix factorization (SNMF), such that the learned features are (a) sparse linear combinations of co-activated neurons, and (b) mapped to their activating inputs, making them directly interpretable. Experiments on Llama 3.1, Gemma 2 and GPT-2 show that SNMF derived features outperform SAEs and a strong supervised baseline (difference-in-means) on causal steering, while aligning with human-interpretable concepts. Further analysis reveals that specific neuron combinations are reused across semantically-related features, exposing a hierarchical structure in the MLP's activation space. Together, these results position SNMF as a simple and effective tool for identifying interpretable features and dissecting concept representations in LLMs.

  • 3 authors
·
Jun 12, 2025 2

SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?

While research on recursive self-improvement (RSI) has predominantly automated model training pipelines, reliable autonomous development demands a missing pillar: post-hoc monitoring and auditing to understand what models learn and ensure safe alignment. Mechanistic interpretability tools are essential to bridge this gap, among which Sparse Autoencoders (SAEs) serve as a cornerstone by isolating interpretable features for model inspection and steering. In this paper, we introduce SAEScientist-Bench to evaluate whether AI agents can act as scientists utilizing SAE tools for autonomous mechanistic discovery. Given a target concept, an agent designs contrastive probes and navigates a Gemma Scope dictionary of 131K+ features in Gemma-2-9B-IT to discover the optimal feature, evaluated against curated expert reference features anchored on Neuronpedia across activation rank, concept selectivity on contrastive texts, and causal steering. Across 10 agent configurations and 20 tasks, frontier agents demonstrate genuine discovery capabilities and lead different evaluation dimensions, but remain well behind the expert baseline, approaching expert levels on separating target concepts from contrastive controls while lagging substantially in causal generation steering. Further analysis reveals that although agents can design contrasts to rule out spurious candidates, they frequently misinterpret experimental measurements. These results establish experimental model understanding as a measurable capability for closed-loop autonomous AI R&D. Our code is available at https://github.com/Trae1ounG/SAEScientist.

Insights into a radiology-specialised multimodal large language model with sparse autoencoders

Interpretability can improve the safety, transparency and trust of AI models, which is especially important in healthcare applications where decisions often carry significant consequences. Mechanistic interpretability, particularly through the use of sparse autoencoders (SAEs), offers a promising approach for uncovering human-interpretable features within large transformer-based models. In this study, we apply Matryoshka-SAE to the radiology-specialised multimodal large language model, MAIRA-2, to interpret its internal representations. Using large-scale automated interpretability of the SAE features, we identify a range of clinically relevant concepts - including medical devices (e.g., line and tube placements, pacemaker presence), pathologies such as pleural effusion and cardiomegaly, longitudinal changes and textual features. We further examine the influence of these features on model behaviour through steering, demonstrating directional control over generations with mixed success. Our results reveal practical and methodological challenges, yet they offer initial insights into the internal concepts learned by MAIRA-2 - marking a step toward deeper mechanistic understanding and interpretability of a radiology-adapted multimodal large language model, and paving the way for improved model transparency. We release the trained SAEs and interpretations: https://huggingface.co/microsoft/maira-2-sae.

  • 6 authors
·
Jul 17, 2025

Hierarchical Sparse Circuit Extraction from Billion-Parameter Language Models through Scalable Attribution Graph Decomposition

Mechanistic interpretability seeks to reverse-engineer neural network computations into human-understandable algorithms, yet extracting sparse computational circuits from billion-parameter language models remains challenging due to exponential search complexity and pervasive polysemanticity. The proposed Hierarchical Attribution Graph Decomposition (HAGD) framework reduces circuit discovery complexity from O(2^n) exhaustive enumeration to O(n^2 log n) through multi-resolution abstraction hierarchies and differentiable circuit search. The methodology integrates cross-layer transcoders for monosemantic feature extraction, graph neural network meta-learning for topology prediction, and causal intervention protocols for validation. Empirical evaluation spans GPT-2 variants, Llama-7B through Llama-70B, and Pythia suite models across algorithmic tasks and natural language benchmarks. On modular arithmetic tasks, the framework achieves up to 91% behavioral preservation (pm2.3\% across runs) while maintaining interpretable subgraph sizes. Cross-architecture transfer experiments suggest that discovered circuits exhibit moderate structural similarity (averaging 67%) across model families, indicating potential shared computational patterns. These results provide preliminary foundations for interpretability at larger model scales while identifying significant limitations in current attribution methodologies that require future advances.

  • 3 authors
·
Jan 19

Activation Steering for Bias Mitigation: An Interpretable Approach to Safer LLMs

As large language models (LLMs) become more integrated into societal systems, the risk of them perpetuating and amplifying harmful biases becomes a critical safety concern. Traditional methods for mitigating bias often rely on data filtering or post-hoc output moderation, which treat the model as an opaque black box. In this work, we introduce a complete, end-to-end system that uses techniques from mechanistic interpretability to both identify and actively mitigate bias directly within a model's internal workings. Our method involves two primary stages. First, we train linear "probes" on the internal activations of a model to detect the latent representations of various biases (e.g., gender, race, age). Our experiments on gpt2-large demonstrate that these probes can identify biased content with near-perfect accuracy, revealing that bias representations become most salient in the model's later layers. Second, we leverage these findings to compute "steering vectors" by contrasting the model's activation patterns for biased and neutral statements. By adding these vectors during inference, we can actively steer the model's generative process away from producing harmful, stereotypical, or biased content in real-time. We demonstrate the efficacy of this activation steering technique, showing that it successfully alters biased completions toward more neutral alternatives. We present our work as a robust and reproducible system that offers a more direct and interpretable approach to building safer and more accountable LLMs.

  • 1 authors
·
Aug 12, 2025

Mechanistic Interpretation through Contextual Decomposition in Transformers

Transformers exhibit impressive capabilities but are often regarded as black boxes due to challenges in understanding the complex nonlinear relationships between features. Interpreting machine learning models is of paramount importance to mitigate risks, and mechanistic interpretability is in particular of current interest as it opens up a window for guiding manual modifications and reverse-engineering solutions. In this work, we introduce contextual decomposition for transformers (CD-T), extending a prior work on CD for RNNs and CNNs, to address mechanistic interpretation computationally efficiently. CD-T is a flexible interpretation method for transformers. It can capture contributions of combinations of input features or source internal components (e.g. attention heads, feed-forward networks) to (1) final predictions or (2) the output of any target internal component. Using CD-T, we propose a novel algorithm for circuit discovery. On a real-world pathology report classification task: we show CD-T distills a more faithful circuit of attention heads with improved computational efficiency (speed up 2x) than a prior benchmark, path patching. As a versatile interpretation method, CD-T also exhibits exceptional capabilities for local interpretations. CD-T is shown to reliably find words and phrases of contrasting sentiment/topic on SST-2 and AGNews datasets. Through human experiments, we demonstrate CD-T enables users to identify the more accurate of two models and to better trust a model's outputs compared to alternative interpretation methods such as SHAP and LIME.

  • 5 authors
·
Jun 30, 2024

DifFRACT: Diffusion Feature Reconstruction and Attribution for Circuit Tracing

Mechanistic interpretability seeks to explain neural network behavior by decomposing model computations into interpretable features and circuits. While transcoder-based circuit tracing has recently enabled detailed causal analyses of large language models, multimodal diffusion transformers for image generation remain comparatively opaque. We still lack tools for understanding how semantic information propagates across denoising steps and how text and image representations interact within double-stream MM-DiT architectures. Existing methods provide only partial insight: attention maps expose a limited view of token interactions, while sparse autoencoders can discover interpretable features but do not directly reveal how these features are transformed and composed through nonlinear MLP layers. In this work, we extend transcoder-based circuit tracing to multimodal diffusion transformers. We train timestep-conditioned transcoders that faithfully approximate the input-output behavior of MLP sublayers in FLUX.1[schnell]. By replacing MLPs with transcoders and linearizing the remaining computation, we obtain exact feature-to-feature attribution and recover compact, interpretable circuits. Empirically, our transcoders match or slightly outperform sparse autoencoders on the sparsity-faithfulness tradeoff. The resulting circuits reveal mechanisms underlying attribute binding and cross-stream semantic propagation, and provide causal explanations for systematic generation errors. Moreover, circuit-guided interventions are substantially more precise and effective than standard SAE-based steering. Our results demonstrate that transcoder-based circuit analysis is feasible for state-of-the-art diffusion transformers and provides a powerful framework for understanding and controlling multimodal generative models. The code is available at https://github.com/Artalmaz31/DifFRACT

  • 3 authors
·
Jun 14

From Residuals to Reasons: LLM-Guided Mechanism Inference from Tabular Data

A persistent challenge in machine learning for scientific applications is jointly achieving prediction and understanding. Statistical models excel on structured data but operate as black boxes, while existing interpretability methods are largely inspective: they answer "which features matter?" but do not articulate how features interact or refine explanations iteratively alongside human understanding. Asking an LLM to predict the target directly forces it to search the entire output space; we instead anchor predictions with a base model and ask the LLM the narrower question of what that model is missing. We introduce Multi-Agent Residual In-Context Learning (MARICL), an agentic framework in which LLM agents analyze where a base-model fails, hypothesize missing structure from high-residual examples provided in context, and produce explicit correction terms refined through multi-turn textual gradient optimization. Across nine benchmarks spanning scientific, biomedical, socioeconomic, and synthetic settings, MARICL improves consistently over its base model on all datasets. To test whether these corrections reflect real structure or batch-specific noise, we freeze formulas learned on one experimental batch of the Cell-Free Protein dataset and apply them (with no retraining and no further LLM calls) to held-out batches. Within the same reagent protocol, the frozen formulas improve predictions in over 92% of cases; across a different protocol, they fail systematically. The success boundary aligns with the biochemistry, not the batch count; direct evidence of mechanistic generalization.

  • 2 authors
·
May 20 1

The Quest for the Right Mediator: A History, Survey, and Theoretical Grounding of Causal Interpretability

Interpretability provides a toolset for understanding how and why neural networks behave in certain ways. However, there is little unity in the field: most studies employ ad-hoc evaluations and do not share theoretical foundations, making it difficult to measure progress and compare the pros and cons of different techniques. Furthermore, while mechanistic understanding is frequently discussed, the basic causal units underlying these mechanisms are often not explicitly defined. In this paper, we propose a perspective on interpretability research grounded in causal mediation analysis. Specifically, we describe the history and current state of interpretability taxonomized according to the types of causal units (mediators) employed, as well as methods used to search over mediators. We discuss the pros and cons of each mediator, providing insights as to when particular kinds of mediators and search methods are most appropriate depending on the goals of a given study. We argue that this framing yields a more cohesive narrative of the field, as well as actionable insights for future work. Specifically, we recommend a focus on discovering new mediators with better trade-offs between human-interpretability and compute-efficiency, and which can uncover more sophisticated abstractions from neural networks than the primarily linear mediators employed in current work. We also argue for more standardized evaluations that enable principled comparisons across mediator types, such that we can better understand when particular causal units are better suited to particular use cases.

  • 13 authors
·
Aug 2, 2024

Bucketing the Good Apples: A Method for Diagnosing and Improving Causal Abstraction

We present a method for diagnosing interpretation in neural networks by identifying an input subspace where a proposed interpretation is highly faithful. Our method is particularly useful for causal-abstraction-style interpretability, where a high-level causal hypothesis is evaluated by interchange interventions. Rather than treating interchange intervention accuracy as a single global summary, we refine this framework by partitioning the input space into well-interpreted and under-interpreted regions according to pairwise interchange-intervention behavior. This turns causal abstraction from a purely global evaluation into a more diagnostic tool: it not only measures whether an interpretation works, but also reveals where it works, where it fails, and what distinguishes the two cases. This diagnostic view also provides practical heuristics for improving interpretations. By analyzing the structure of the well-interpreted and under-interpreted regions, we can identify missing distinctions in a high-level hypothesis, discover previously unmodeled intermediate variables, and combine complementary partial interpretations into a stronger one. We instantiate this idea as a simple four-step recipe and show that it yields informative error analyses across multiple causal abstraction settings. In a toy logic task, recursively applying the recipe recovers a high-level hypothesis from scratch. More broadly, our results suggest that partitioning the input space is a useful step toward more precise, constructive, and scalable mechanistic interpretability.

  • 5 authors
·
May 3

Clinical Concept Centers in LLMs

Large language models are increasingly used in clinical settings. However, research into the reliability and performance of these models has focused almost entirely on the language substrate, scoring what the model says. Mechanistic interpretability has found that the latent space carries a higher fidelity of representation than the text: internal representations not only encode substantially more than the output verbalizes, but the stated reasoning also systematically omits features that causally drive the answer. An evaluation of model behavior in terms of mechanistic interpretability has not been explored in clinical decision support. In this work, we extend behavioral evaluation into the latent space and ask whether clinical concepts exist as locatable, causally used representations inside open-weight LLMs. We find dedicated clinical concept centers in the latent space of all eleven open models we test. These concept centers are interpretable, firing only on their aligned clinical narratives, and meaningfully and causally drive model behavior in both constrained and open-ended settings. They are not just analytical representations, but circuits that can be utilized in clinical practice, and we explore their use from the perspective of both evaluation and performance. From the evaluation standpoint, models stay internally coherent and keep using the relevant concept centers even under adversarial role-based priming, while aligned priming improves downstream clinical performance. From a performance perspective, we simulate realistic deployment settings and find that steering models along these centers leads to meaningful downstream improvements. Finally, we conduct a blinded clinician validation and find the activation and usage of these concept centers predicts clinicians preferences.

  • 5 authors
·
Oct 4

Sparse Autoencoders Do Not Find Canonical Units of Analysis

A common goal of mechanistic interpretability is to decompose the activations of neural networks into features: interpretable properties of the input computed by the model. Sparse autoencoders (SAEs) are a popular method for finding these features in LLMs, and it has been postulated that they can be used to find a canonical set of units: a unique and complete list of atomic features. We cast doubt on this belief using two novel techniques: SAE stitching to show they are incomplete, and meta-SAEs to show they are not atomic. SAE stitching involves inserting or swapping latents from a larger SAE into a smaller one. Latents from the larger SAE can be divided into two categories: novel latents, which improve performance when added to the smaller SAE, indicating they capture novel information, and reconstruction latents, which can replace corresponding latents in the smaller SAE that have similar behavior. The existence of novel features indicates incompleteness of smaller SAEs. Using meta-SAEs -- SAEs trained on the decoder matrix of another SAE -- we find that latents in SAEs often decompose into combinations of latents from a smaller SAE, showing that larger SAE latents are not atomic. The resulting decompositions are often interpretable; e.g. a latent representing ``Einstein'' decomposes into ``scientist'', ``Germany'', and ``famous person''. Even if SAEs do not find canonical units of analysis, they may still be useful tools. We suggest that future research should either pursue different approaches for identifying such units, or pragmatically choose the SAE size suited to their task. We provide an interactive dashboard to explore meta-SAEs: https://metasaes.streamlit.app/

  • 8 authors
·
Feb 6, 2025

CodeCircuit: Toward Inferring LLM-Generated Code Correctness via Attribution Graphs

Current paradigms for code verification rely heavily on external mechanisms-such as execution-based unit tests or auxiliary LLM judges-which are often labor-intensive or limited by the judging model's own capabilities. This raises a fundamental, yet unexplored question: Can an LLM's functional correctness be assessed purely from its internal computational structure? Our primary objective is to investigate whether the model's neural dynamics encode internally decodable signals that are predictive of logical validity during code generation. Inspired by mechanistic interpretability, we propose to treat code verification as a mechanistic diagnostic task, mapping the model's explicit algorithmic trajectory into line-level attribution graphs. By decomposing complex residual flows, we aim to identify the structural signatures that distinguish sound reasoning from logical failure within the model's internal circuits. Analysis across Python, C++, and Java confirms that intrinsic correctness signals are robust across diverse syntaxes. Topological features from these internal graphs predict correctness more reliably than surface heuristics and enable targeted causal interventions to fix erroneous logic. These findings establish internal introspection as a decodable property for verifying generated code. Our code is at https:// github.com/bruno686/CodeCircuit.

Certified Circuits: Stability Guarantees for Mechanistic Circuits

Understanding how neural networks arrive at their predictions is essential for debugging, auditing, and deployment. Mechanistic interpretability pursues this goal by identifying circuits--minimal subnetworks responsible for specific behaviors. However, existing circuit discovery methods are brittle: circuits depend strongly on the chosen concept dataset and often fail to transfer out-of-distribution, raising doubts whether they capture the concept or merely dataset-specific artifacts. We introduce Certified Circuits, which provide provable stability guarantees for circuit discovery. Our framework wraps any black-box discovery algorithm with randomized data subsampling to certify that inclusion decisions over circuit components--neurons or edges of the model graph, depending on the base algorithm--are invariant to bounded edit-distance perturbations of the concept dataset. Unstable components are abstained from, yielding circuits that are more compact and more accurate. We validate across three architectures (ResNet, ViT, GPT-2) on vision (ImageNet and four OOD datasets) and language (IOI, IOI-Hard, Greater-Than) tasks. Certified circuits achieve up to 56% higher accuracy and up to 80% fewer components, and remain reliable where baselines degrade. Certified Circuits puts circuit discovery on formal ground by producing mechanistic explanations that are provably stable and better aligned with the target concept. Code: https://github.com/AlaaAnani/certified-circuits.

  • 5 authors
·
May 27

U-Space: Uncovering When and Why Uncertainty Arises in Language Models

Large language models are informing decisions with ever-higher stakes. As the consequences of their errors grow, a central question becomes harder to ignore: how much can we trust an individual answer? Yet recognizing when to defer remains difficult because language models can present incorrect conclusions with fluent explanations and an authoritative tone. Uncertainty quantification seeks to address this disconnect by estimating the reliability of individual predictions. However, many existing methods require repeated generations or separately trained components, and their scalar estimates do not reveal where uncertainty arises or how it evolves during reasoning. Recent work has also shown that generation length can be strongly associated with uncertainty estimates and correctness, raising the question of how much of an estimator's predictive power comes from uncertainty-specific information rather than output length alone. Mechanistic interpretability offers a way to address these limitations by connecting human-interpretable concepts to intermediate model states. Building on this capability, we introduce the U-Space, a low-dimensional subspace that makes a model's evolving uncertainty measurable and interpretable. We identify semantic anchors for doubt and certainty, map their unembedding directions back into the residual space, and combine their contrasts into an orthogonal basis. The U-Lens projects each token state onto these basis vectors, yielding an interpretable token-level uncertainty map that can be inspected directly or aggregated into a scalar uncertainty score. Our approach requires no correctness labels, repeated generations, or training. Across reasoning benchmarks, its confidence score outperforms established baselines under both standard and length-controlled evaluation and transfers more reliably than supervised estimators. Code: https://github.com/s2labres/U-Space.

EVA: Towards a universal model of the immune system

The effective application of foundation models to translational research in immune-mediated diseases requires multimodal patient-level representations that can capture complex phenotypes emerging from multicellular interactions. Yet most current biological foundation models focus only on single-cell resolution and are evaluated on technical metrics often disconnected from actual drug development tasks and challenges. Here, we introduce EVA, the first cross-species, multimodal foundation model of immunology and inflammation, a therapeutic area where shared pathogenic mechanisms create unique opportunities for transfer learning. EVA harmonizes transcriptomics data across species, platforms, and resolutions, and integrates histology data to produce rich, unified patient representations. We establish clear scaling laws, demonstrating that increasing model size and compute translates to improvements in both pretraining and downstream tasks performance. We introduce a comprehensive evaluation suite of 39 tasks spanning the drug development pipeline: zero-shot target efficacy and gene function prediction for discovery, cross-species or cross-diseases molecular perturbations for preclinical development, and patient stratification with treatment response prediction or disease activity prediction for clinical trials applications. We benchmark EVA against several state-of-the-art biological foundation models and baselines on these tasks, and demonstrate state-of-the-art results on each task category. Using mechanistic interpretability, we further identify biological meaningful features, revealing intertwined representations across species and technologies. We release an open version of EVA for transcriptomics to accelerate research on immune-mediated diseases.

  • 11 authors
·
Feb 10

Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models

Predicting the answer to interventional ``what if'' questions --- the outcome of an action never taken --- requires a mechanistic, causal model, not a curve fit; and learning such a model requires experiments, because passive data leaves its mechanisms unidentified. Experiments are expensive, so the central problem is data efficiency. We present the Model Discovery Agent (MDA), which couples a large language model (LLM), used as a proposer of candidate structures, with standard Bayesian machinery --- sequential Monte Carlo (SMC) for parameter and structure posteriors, simulation-based inference (SBI) for intractable likelihoods, and value-of-information (VoI) for experiment design --- to discover latent mechanistic world models from few interventions. MDA operates in the M-open setting: when the truth lies outside the current hypothesis class, a predictive check flags the inadequacy and the proposer expands the hypothesis space with a new model whose parameters are then identified by designed experiments. We show that discovery and design reinforce: the design step identifies the mechanism the discovery step proposes, and the identified mechanism improves predictions, enabling further discoveries from the remaining unexplained residuals. On three different benchmarks --- covering physics (\DPbench, wiemann2026discoverphysics), chemistry (\CHEMbench, kabra2026autoscilab) and biology (\HHbench, a new partially observed single-neuron electrophysiology benchmark we create) --- we show that MDA sets a new SOTA in terms of data-efficient model learning and reliable interventional prediction ability.

  • 1 authors
·
Aug 12

When the Coffee Feature Activates on Coffins: An Analysis of Feature Extraction and Steering for Mechanistic Interpretability

Recent work by Anthropic on Mechanistic interpretability claims to understand and control Large Language Models by extracting human-interpretable features from their neural activation patterns using sparse autoencoders (SAEs). If successful, this approach offers one of the most promising routes for human oversight in AI safety. We conduct an initial stress-test of these claims by replicating their main results with open-source SAEs for Llama 3.1. While we successfully reproduce basic feature extraction and steering capabilities, our investigation suggests that major caution is warranted regarding the generalizability of these claims. We find that feature steering exhibits substantial fragility, with sensitivity to layer selection, steering magnitude, and context. We observe non-standard activation behavior and demonstrate the difficulty to distinguish thematically similar features from one another. While SAE-based interpretability produces compelling demonstrations in selected cases, current methods often fall short of the systematic reliability required for safety-critical applications. This suggests a necessary shift in focus from prioritizing interpretability of internal representations toward reliable prediction and control of model output. Our work contributes to a more nuanced understanding of what mechanistic interpretability has achieved and highlights fundamental challenges for AI safety that remain unresolved.

  • 3 authors
·
Jan 6

Transcoders Find Interpretable LLM Feature Circuits

A key goal in mechanistic interpretability is circuit analysis: finding sparse subgraphs of models corresponding to specific behaviors or capabilities. However, MLP sublayers make fine-grained circuit analysis on transformer-based language models difficult. In particular, interpretable features -- such as those found by sparse autoencoders (SAEs) -- are typically linear combinations of extremely many neurons, each with its own nonlinearity to account for. Circuit analysis in this setting thus either yields intractably large circuits or fails to disentangle local and global behavior. To address this we explore transcoders, which seek to faithfully approximate a densely activating MLP layer with a wider, sparsely-activating MLP layer. We successfully train transcoders on language models with 120M, 410M, and 1.4B parameters, and find them to perform at least on par with SAEs in terms of sparsity, faithfulness, and human-interpretability. We then introduce a novel method for using transcoders to perform weights-based circuit analysis through MLP sublayers. The resulting circuits neatly factorize into input-dependent and input-invariant terms. Finally, we apply transcoders to reverse-engineer unknown circuits in the model, and we obtain novel insights regarding the greater-than circuit in GPT2-small. Our results suggest that transcoders can prove effective in decomposing model computations involving MLPs into interpretable circuits. Code is available at https://github.com/jacobdunefsky/transcoder_circuits.

  • 3 authors
·
Jun 17, 2024

Gradients for Interventions and Activations for Detection: Targeted Feature Learning in Language Models

Model-internal features can be studied through both their ability to identify a specified concept and their causal effect when manipulated, e.g., through steering or weight editing. A prominent approach to feature learning is Sparse Autoencoders (SAEs), which learn broad feature dictionaries whose relation to particular concepts is typically identified post hoc. However, many interpretability questions are instead hypothesis-driven and concern a concept specified in advance. We study this setting as targeted feature learning, where a single feature is constructed for such a predefined concept. We present a controlled comparison across three model signals (activation values, activation gradients, and parameter gradients) and two estimators (contrastive mean and a learned one-dimensional encoder-decoder), yielding six targeted methods, with CAA and GRADIEND as existing instances and four new methods covering the remaining combinations. We compare these methods against pretrained SAEs across 15 tasks and three language models, evaluating both detection and causal intervention. Across models, the strongest detection performance is achieved by contrastive activation value methods, whereas the strongest intervention performance is achieved by gradient-based methods. Overall, our results show that targeted feature quality depends jointly on the model signal and estimator, with detection and intervention capturing complementary properties.

  • 2 authors
·
Sep 25

Matryoshka attribution: Learning to attribute language model outputs to representations and weights

Attributing language model outputs to their internal computations is an open problem in interpretability. Existing methods, which use causal interventions, gradients, or learnable masks, either are infeasibly expensive or struggle to identify actual causally-important internal computations. We propose framing attribution as the problem of identifying nested subsets of internal components which minimise a downstream loss. To learn this task, we introduce Matryoshka Attribution (MAttr), a mask learning method that parametrises the mask with a simple differentiable sigmoid top-k operator. We supervise training over all sparsities simultaneously by randomising k over training, resulting in a learned ordering of components by attribution score. MAttr achieves number 1 on the official leaderboard of the Mechanistic Interpretability Benchmark (Mueller et al., 2025); our method identifies sparse and task-transferrable circuits across varying circuit bases. As a practical application, we show that MAttr can be trained with reinforcement learning to identify weight changes responsible for downstream behaviours in LLM finetuning. We train MAttr on refusal judge scores and find that restoring 1% of Llama 3.1 8B Instruct's weights to their base model state is sufficient to remove refusals while maintaining capabilities. We view MAttr as a successful formulation of interpretability into a learnable objective that we can tackle with gradient descent, and encourage future work along these lines.

  • 7 authors
·
Sep 21

Glyph: An Interpretability Engine for AI-Driven Scientific Discovery in Clinical Data

Automated scientific discovery has the potential to advance understanding of disease mechanisms and accelerate identification of early disease markers. Manual data annotation is infeasible at population scale, motivating methods that automatically extract pre-defined phenotypes in multimodal and unstructured clinical data. Such pre-defined feature sets determine and limit which associations these approaches can discover. This limitation extends to agentic AI systems that use these tools to query population-scale data. In contrast, foundation models (FMs) trained on large volumes of clinical data can perform tasks beyond their training objectives. This suggests their latent embeddings may represent novel or unexplored features that could yield insights into disease mechanisms, yet such latent features remain difficult to interpret. To address this challenge, we introduce Glyph, an automated multistage framework that extracts features from multimodal and unstructured clinical data using FMs and sparse autoencoders, interprets them as coded clinical concepts, and evaluates their value for clinical scientific discovery. We applied Glyph to abdominal CT volumes, CT radiology reports and laboratory measurements from 41,628 patients across different institutions. We identified 397 predictive concepts supported by 4,463 Bonferroni-significant associations. Moreover, a majority of discovered associations were automatically reproducible in held-out cohorts across different sites. Glyph concepts provided complementary associations to literature-derived markers, leading to gains in predictive power. We envision that interpretability can extend beyond a mechanistic understanding of neural networks to connecting implicitly learned features with clinical knowledge and establishing their significance for patient outcomes. We release Glyph at https://github.com/StanfordMIMI/Glyph.

Sparsity is All You Need: Rethinking Biological Pathway-Informed Approaches in Deep Learning

Biologically-informed neural networks typically leverage pathway annotations to enhance performance in biomedical applications. We hypothesized that the benefits of pathway integration does not arise from its biological relevance, but rather from the sparsity it introduces. We conducted a comprehensive analysis of all relevant pathway-based neural network models for predictive tasks, critically evaluating each study's contributions. From this review, we curated a subset of methods for which the source code was publicly available. The comparison of the biologically informed state-of-the-art deep learning models and their randomized counterparts showed that models based on randomized information performed equally well as biologically informed ones across different metrics and datasets. Notably, in 3 out of the 15 analyzed models, the randomized versions even outperformed their biologically informed counterparts. Moreover, pathway-informed models did not show any clear advantage in interpretability, as randomized models were still able to identify relevant disease biomarkers despite lacking explicit pathway information. Our findings suggest that pathway annotations may be too noisy or inadequately explored by current methods. Therefore, we propose a methodology that can be applied to different domains and can serve as a robust benchmark for systematically comparing novel pathway-informed models against their randomized counterparts. This approach enables researchers to rigorously determine whether observed performance improvements can be attributed to biological insights.

  • 7 authors
·
May 6, 2025

From Data Statistics to Feature Geometry: How Correlations Shape Superposition

A central idea in mechanistic interpretability is that neural networks represent more features than they have dimensions, arranging them in superposition to form an over-complete basis. This framing has been influential, motivating dictionary learning approaches such as sparse autoencoders. However, superposition has mostly been studied in idealized settings where features are sparse and uncorrelated. In these settings, superposition is typically understood as introducing interference that must be minimized geometrically and filtered out by non-linearities such as ReLUs, yielding local structures like regular polytopes. We show that this account is incomplete for realistic data by introducing Bag-of-Words Superposition (BOWS), a controlled setting to encode binary bag-of-words representations of internet text in superposition. Using BOWS, we find that when features are correlated, interference can be constructive rather than just noise to be filtered out. This is achieved by arranging features according to their co-activation patterns, making interference between active features constructive, while still using ReLUs to avoid false positives. We show that this kind of arrangement is more prevalent in models trained with weight decay and naturally gives rise to semantic clusters and cyclical structures which have been observed in real language models yet were not explained by the standard picture of superposition. Code for this paper can be found at https://github.com/LucasPrietoAl/correlations-feature-geometry.

  • 5 authors
·
Mar 9

Momentum Attention: The Physics of In-Context Learning and Spectral Forensics for Mechanistic Interpretability

The Mechanistic Interpretability (MI) program has mapped the Transformer as a precise computational graph. We extend this graph with a conservation law and time-varying AC dynamics, viewing it as a physical circuit. We introduce Momentum Attention, a symplectic augmentation embedding physical priors via the kinematic difference operator p_t = q_t - q_{t-1}, implementing the symplectic shear q_t = q_t + γp_t on queries and keys. We identify a fundamental Symplectic-Filter Duality: the physical shear is mathematically equivalent to a High-Pass Filter. This duality is our cornerstone contribution -- by injecting kinematic momentum, we sidestep the topological depth constraint (L geq 2) for induction head formation. While standard architectures require two layers for induction from static positions, our extension grants direct access to velocity, enabling Single-Layer Induction and Spectral Forensics via Bode Plots. We formalize an Orthogonality Theorem proving that DC (semantic) and AC (mechanistic) signals segregate into orthogonal frequency bands when Low-Pass RoPE interacts with High-Pass Momentum. Validated through 5,100+ controlled experiments (documented in Supplementary Appendices A--R and 27 Jupyter notebooks), our 125M Momentum model exceeds expectations on induction-heavy tasks while tracking a 350M baseline within sim2.9% validation loss. Dedicated associative recall experiments reveal a scaling law γ^* = 4.17 times N^{-0.74} establishing momentum-depth fungibility. We offer this framework as a complementary analytical toolkit connecting Generative AI, Hamiltonian Physics, and Signal Processing.

  • 1 authors
·
Feb 3

A Function Interpretation Benchmark for Evaluating Interpretability Methods

Labeling neural network submodules with human-legible descriptions is useful for many downstream tasks: such descriptions can surface failures, guide interventions, and perhaps even explain important model behaviors. To date, most mechanistic descriptions of trained networks have involved small models, narrowly delimited phenomena, and large amounts of human labor. Labeling all human-interpretable sub-computations in models of increasing size and complexity will almost certainly require tools that can generate and validate descriptions automatically. Recently, techniques that use learned models in-the-loop for labeling have begun to gain traction, but methods for evaluating their efficacy are limited and ad-hoc. How should we validate and compare open-ended labeling tools? This paper introduces FIND (Function INterpretation and Description), a benchmark suite for evaluating the building blocks of automated interpretability methods. FIND contains functions that resemble components of trained neural networks, and accompanying descriptions of the kind we seek to generate. The functions are procedurally constructed across textual and numeric domains, and involve a range of real-world complexities, including noise, composition, approximation, and bias. We evaluate new and existing methods that use language models (LMs) to produce code-based and language descriptions of function behavior. We find that an off-the-shelf LM augmented with only black-box access to functions can sometimes infer their structure, acting as a scientist by forming hypotheses, proposing experiments, and updating descriptions in light of new data. However, LM-based descriptions tend to capture global function behavior and miss local corruptions. These results show that FIND will be useful for characterizing the performance of more sophisticated interpretability methods before they are applied to real-world models.

  • 8 authors
·
Sep 7, 2023

Don't Judge an LLM Only by Its Activations: Discovering Suppressed Safety Features via Counterfactual Activation Potential

Mechanistic interpretability has emerged as the primary means to understand safety behavior of LLMs. However, existing tools primarily focus on the activating neurons or features of a model. The role of the remaining large set of inactive components is invisible to such methods. This work demonstrates that the inactive set contains safety-critical features that are causally relevant for refusal of harmful prompts. Suppressing such features could turn refusals into compliance, while passing undetected by prevalent interpretability tools. We introduce the Counterfactual Activation Potential (CAP), a metric that quantifies a suppressed feature's latent activation tendency as the product of its encoder alignment (how strongly the input drives it), suppression strength (how strongly active features inhibit it), and safety criticality (how much refusal depends on it). To find suppressed safety features at scale, we propose CAP-guided Safety Feature Discovery (CSFD), a two-stage filtering algorithm that identifies candidate safety features from hundreds of thousands of transcoder features without exhaustive ablation. A significant fraction of trials turn compliant with harmful prompts when a candidate feature is ablated. Under natural jailbreaks, the suppression acting on the highest-CAP features rises 2-4x, and their activation correspondingly falls by up to 80%. Amplifying a feature's suppressors pushes its activation down and raises harmful compliance with prompts related to the suppressed feature, with no such effect for random features. Our experiments span five Gemma, Qwen, and Llama models across various parameter sizes. Our findings indicate that jailbreaks could operate in part by suppressing safety-critical features rather than solely activating harmful ones, and that suppressed features are a necessary complement to activation-focused interpretability of safety behavior.

  • 2 authors
·
Oct 3

Simplifying Outcomes of Language Model Component Analyses with ELIA

While mechanistic interpretability has developed powerful tools to analyze the internal workings of Large Language Models (LLMs), their complexity has created an accessibility gap, limiting their use to specialists. We address this challenge by designing, building, and evaluating ELIA (Explainable Language Interpretability Analysis), an interactive web application that simplifies the outcomes of various language model component analyses for a broader audience. The system integrates three key techniques -- Attribution Analysis, Function Vector Analysis, and Circuit Tracing -- and introduces a novel methodology: using a vision-language model to automatically generate natural language explanations (NLEs) for the complex visualizations produced by these methods. The effectiveness of this approach was empirically validated through a mixed-methods user study, which revealed a clear preference for interactive, explorable interfaces over simpler, static visualizations. A key finding was that the AI-powered explanations helped bridge the knowledge gap for non-experts; a statistical analysis showed no significant correlation between a user's prior LLM experience and their comprehension scores, suggesting that the system reduced barriers to comprehension across experience levels. We conclude that an AI system can indeed simplify complex model analyses, but its true power is unlocked when paired with thoughtful, user-centered design that prioritizes interactivity, specificity, and narrative guidance.

  • 2 authors
·
Feb 20

The Non-Linear Representation Dilemma: Is Causal Abstraction Enough for Mechanistic Interpretability?

The concept of causal abstraction got recently popularised to demystify the opaque decision-making processes of machine learning models; in short, a neural network can be abstracted as a higher-level algorithm if there exists a function which allows us to map between them. Notably, most interpretability papers implement these maps as linear functions, motivated by the linear representation hypothesis: the idea that features are encoded linearly in a model's representations. However, this linearity constraint is not required by the definition of causal abstraction. In this work, we critically examine the concept of causal abstraction by considering arbitrarily powerful alignment maps. In particular, we prove that under reasonable assumptions, any neural network can be mapped to any algorithm, rendering this unrestricted notion of causal abstraction trivial and uninformative. We complement these theoretical findings with empirical evidence, demonstrating that it is possible to perfectly map models to algorithms even when these models are incapable of solving the actual task; e.g., on an experiment using randomly initialised language models, our alignment maps reach 100% interchange-intervention accuracy on the indirect object identification task. This raises the non-linear representation dilemma: if we lift the linearity constraint imposed to alignment maps in causal abstraction analyses, we are left with no principled way to balance the inherent trade-off between these maps' complexity and accuracy. Together, these results suggest an answer to our title's question: causal abstraction is not enough for mechanistic interpretability, as it becomes vacuous without assumptions about how models encode information. Studying the connection between this information-encoding assumption and causal abstraction should lead to exciting future work.

  • 4 authors
·
Jul 11, 2025

How Do Agentic LLMs Decide to Call Tools? A Tool-Call Vector Shaped by Suppression

Tool calling, invoking external tools on demand, is central to agentic LLMs, yet the mechanism that decides whether a model calls a tool or responds directly remains poorly understood. Agentic prompts are long and heavily scaffolded, combining role instructions, tool schemas, format templates, and the user's request across hundreds of tokens, creating a noisy, highly entangled context in which no single controllable variable for mechanistic analysis is obvious. To obtain such a variable, we propose a method that converts complex agentic prompts into minimal contrastive pairs in which a single request verb determines the tool-call decision: replacing an execution-verb (e.g., write) with an analysis-verb (e.g., discuss) reliably flips the decision, suggesting it is mediated by a compact internal state. We construct 500 such paired prompts across Python, Java, and C++ (300 for mechanistic analysis, 200 held out for evaluation). We trace the decision to a vector, μ_Δ, that is both causally necessary and sufficient and generalizes beyond the discovery prompts to native multi-turn τ^2-Bench trajectories and verb-free requests. Behavioral ablations show that the scaffold establishes a tool-call prior; Transcoder decomposition then reveals that analysis verbs suppress this prior through features signaling that tool use is unnecessary, whereas execution verbs largely leave it intact. Downstream scaffold-reading attention heads and MLP features read out the resulting state, and the same mechanism recurs across seven models from the Qwen, Mistral, and Granite families. Our code is available at https://github.com/XijieGo/MI4ToolCalling.

  • 8 authors
·
Oct 6

Induction Signatures Are Not Enough: A Matched-Compute Study of Load-Bearing Structure in In-Context Learning

Mechanism-targeted synthetic data is increasingly proposed as a way to steer pretraining toward desirable capabilities, but it remains unclear how such interventions should be evaluated. We study this question for in-context learning (ICL) under matched compute (iso-FLOPs) using Bi-Induct, a lightweight data rewrite that interleaves short directional copy snippets into a natural pretraining stream: forward-copy (induction), backward-copy (anti-induction, as a directional control), or a balanced mix. Across 0.13B-1B decoder-only models, we evaluate (i) few-shot performance on standard LM benchmarks and function-style ICL probes, (ii) head-level copy telemetry, and (iii) held-out perplexity as a guardrail. Bi-Induct reliably increases induction-head activity, but this does not translate into consistent improvements in few-shot generalization: on standard LM benchmarks, Bi-Induct is largely performance-neutral relative to natural-only training, while on function-style probes the 1B natural-only model performs best. Despite explicit backward-copy cues, anti-induction scores remain near zero across scales, revealing a strong forward/backward asymmetry. Targeted ablations show a sharper distinction: removing the top 2% induction heads per layer harms ICL more than matched random ablations, with the largest relative drop occurring in the natural-only models. This indicates that natural-only training produces more centralized, load-bearing induction circuitry, whereas Bi-Induct tends to create more distributed and redundant induction activity. Our main conclusion is that eliciting a mechanism is not the same as making it load-bearing. For data-centric foundation model design, this suggests that synthetic data interventions should be evaluated not only by signature amplification, but by whether they create causally necessary computation while preserving natural-data modeling quality.

  • 2 authors
·
Mar 13

Qwen-Scope: Turning Sparse Features into Development Tools for Large Language Models

Large language models have achieved remarkable capabilities across diverse tasks, yet their internal decision-making processes remain largely opaque, limiting our ability to inspect, control, and systematically improve them. This opacity motivates a growing body of research in mechanistic interpretability, with sparse autoencoders (SAEs) emerging as one of the most promising tools for decomposing model activations into sparse, interpretable feature representations. We introduce Qwen-Scope, an open-source suite of SAEs built on the Qwen model family, comprising 14 groups of SAEs across 7 model variants from the Qwen3 and Qwen3.5 series, covering both dense and mixture-of-expert architectures. Built on top of these SAEs, we show that SAEs can go beyond post-hoc analysis to serve as practical interfaces for model development along four directions: (i) inference-time steering, where SAE feature directions control language, concepts, and preferences without modifying model weights; (ii) evaluation analysis, where activated SAE features provide a representation-level proxy for benchmark redundancy and capability coverage; (iii) data-centric workflows, where SAE features support multilingual toxicity classification and safety-oriented data synthesis; and (iv) post-training optimization, where SAE-derived signals are incorporated into supervised fine-tuning and reinforcement learning objectives to mitigate undesirable behaviors such as code-switching and repetition. Together, these results demonstrate that SAEs can serve not only as post-hoc analysis tools, but also as reusable representation-level interfaces for diagnosing, controlling, evaluating, and improving large language models. By open-sourcing Qwen-Scope, we aim to support mechanistic research and accelerate practical workflows that connect model internals to downstream behavior.

  • 18 authors
·
May 11

Tool-Augmented On-Policy Distillation for LLM Domain Adaptation in Sequence-Based Omics Tasks

Multi-omics sequences contain complex biological patterns, yet deciphering their mechanisms for automated scientific discovery remains challenging. As large language models (LLMs) interpret these sequences, evaluating both predictions and scientific reasoning is critical. However, existing benchmarks for multi-omics sequence tasks rely on classification and regression metrics, neglecting whether models grasp the underlying biological evidence. We introduce OmicsBench, the first reasoning benchmark for multi-omics sequences, comprising 1,160 expert-validated questions across six tasks spanning DNA regulation, RNA processing, and protein function. OmicsBench requires traceable evidence chains, evaluated using instance-specific rubrics developed with domain experts. Evaluating 17 LLMs reveals an inverse relationship: while scientific LLMs outperform general-purpose LLMs in sequence classification accuracy, they fail to provide valid evidence to support their predictions. One plausible interpretation is shortcut learning: specialized models may rely on statistical patterns rather than the biological mechanisms needed for scientific discovery. Motivated by this finding, we introduce tool-augmented on-policy distillation (TA-OPD), a post-training method to align sequence prediction with evidence-grounded biological reasoning. Across five Qwen3.5 models spanning 0.8B to 27B parameters, TA-OPD consistently strengthens biological evidence grounding while improving predictive performance on most tasks. These gains persist across model scales, indicating that stronger sequence reasoning does not arise solely from increased model capacity, but can be improved through evidence-aware training. Together, OmicsBench and TA-OPD provide a framework for diagnosing reasoning failures in multi-omics LLMs and a path toward models whose predictions are better grounded in biologically meaningful evidence.

  • 14 authors
·
Sep 19

dIon: Fragmentation-Based Invariance for Self-Supervised Learning of Tandem Mass Spectra

We introduce a novel invariance for peptide tandem mass spectrometry data, unlocking self-supervised representation learning that improves de novo sequencing of peptides. This invariance exploits the physical relationship between precursor properties (mass and charge) and fragment-ion evidence, without requiring peptide sequence labels. We introduce dIon, which adapts the DINO framework with two latent prediction tasks, both recovering a clean teacher representation: one from a spectrum mixture, using the precursor as a selection query, and one from a partial spectrum with the precursor withheld. The first associates precursor information with fragment-ion evidence; the second prevents representational collapse onto that information alone. Mechanistic probes support both effects, and ablations show that the full objective performs best. Under identical end-to-end training, dIon initialization improves de novo peptide precision over training from scratch by 5.5 and 8.4 percentage points on the held-out MassIVE-KB and Kingdoms test sets, and by 2.3 and 4.8 percentage points with a larger supervised training corpus. The resulting models surpass fully supervised state-of-the-art de novo sequencing models on the diverse, multi-species Kingdoms corpus under the same greedy-decoding protocol. Without peptide labels, dIon learns strong native peptide-similarity geometry compared with other learned models; with limited peptide-supervised adaptation, it achieves the best retrieval and pair-discrimination performance across all representation benchmarks.

  • 5 authors
·
Oct 4

GIST: Targeted Data Selection for Instruction Tuning via Coupled Optimization Geometry

Targeted data selection has emerged as a crucial paradigm for efficient instruction tuning, aiming to identify a small yet influential subset of training examples for a specific target task. In practice, influence is often measured through the effect of an example on parameter updates. To make selection scalable, many approaches leverage optimizer statistics (e.g., Adam states) as an axis-aligned surrogate for update geometry (i.e., diagonal precondition), implicitly treating parameters as coordinate-wise independent. We show that this assumption breaks down in parameter-efficient fine-tuning (PEFT) methods such as LoRA. In this setting, the induced optimization geometry exhibits strong cross-parameter coupling with non-trivial off-diagonal interactions, while the task-relevant update directions are confined to a low-dimensional subspace. Motivated by this mismatch, we propose GIST (Gradient Isometric Subspace Transformation), a simple yet principled alternative that replaces axis-aligned scaling with robust subspace alignment. GIST recovers a task-specific subspace from validation gradients via spectral filtering (SVD), projects training gradients into this coupled subspace, and scores examples by their alignment with target directions.Extensive experiments have demonstrated that GIST matches or outperforms the state-of-the-art baseline with only 0.29% of the storage and 25% of the computational time under the same selection budget.

Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal

Language-model post-training is the main stage at which model behavior is shaped, yet it still largely involves optimization of scalar rewards that summarize diverse desiderata. This abstraction gives practitioners little visibility into what their data actually teaches models, allowing spurious correlations to be learned by a model and inducing undesirable behaviors such as over-stylization and sycophancy. To address this problem, we ask: can we inspect a preference dataset before optimization and decide, at the level of concepts, which behaviors a model should be allowed to learn? Motivated by this, we introduce a data-centric post-training pipeline that uses interpretability protocols to develop statistical hypotheses for the latent concepts separating preferred from dispreferred generations, making them explicit for fine-grained user feedback. Building on this view, we unify several interpretability-based training protocols as ways of shaping rewards via feature or data interventions. Empirically, we show that our pipeline diagnoses undesirable signals in existing preference data, mitigates off-target learning, and can also help amplify or shape desired properties such as safeguards and model personality. More broadly, our results suggest that interpretability can turn post-training from optimizing opaque proxy rewards into a process of auditing and sculpting the learning signal itself.

  • 17 authors
·
Jun 10

Physics Steering: Causal Control of Cross-Domain Concepts in a Physics Foundation Model

Recent advances in mechanistic interpretability have revealed that large language models (LLMs) develop internal representations corresponding not only to concrete entities but also distinct, human-understandable abstract concepts and behaviour. Moreover, these hidden features can be directly manipulated to steer model behaviour. However, it remains an open question whether this phenomenon is unique to models trained on inherently structured data (ie. language, images) or if it is a general property of foundation models. In this work, we investigate the internal representations of a large physics-focused foundation model. Inspired by recent work identifying single directions in activation space for complex behaviours in LLMs, we extract activation vectors from the model during forward passes over simulation datasets for different physical regimes. We then compute "delta" representations between the two regimes. These delta tensors act as concept directions in activation space, encoding specific physical features. By injecting these concept directions back into the model during inference, we can steer its predictions, demonstrating causal control over physical behaviours, such as inducing or removing some particular physical feature from a simulation. These results suggest that scientific foundation models learn generalised representations of physical principles. They do not merely rely on superficial correlations and patterns in the simulations. Our findings open new avenues for understanding and controlling scientific foundation models and has implications for AI-enabled scientific discovery.

  • 5 authors
·
Nov 25, 2025

MOSAIC: Module Discovery via Sparse Additive Identifiable Causal Learning for Scientific Time Series

Causal representation learning (CRL) seeks to recover latent variables with identifiability guarantees, typically up to permutation and component-wise reparameterization under appropriate assumptions. However, identifiability does not imply interpretability: latent semantics are typically assigned post hoc by alignment with known ground-truth factors. This limitation is particularly acute in scientific time series, where underlying mechanisms are unknown and discovering interpretable structure is a primary goal. In contrast, scientific observations (such as residue-pair distances, climate indices, or process sensors) are inherently semantic, as they correspond to named physical quantities. This raises a key question: can the interpretability of observations be transferred to the identifiable latent space? We propose MOSAIC (Module discovery via Sparse Additive Identifiable Causal learning), a sparse temporal VAE that integrates temporal CRL identifiability with support recovery over observed variables. MOSAIC identifies latent variables via regime-conditioned temporal variation, and recovers for each latent a sparse set of associated observations through an additive decoder, yielding module-level interpretability. We show that ANOVA main-effect supports are identifiable under general smooth mixing functions, and provide finite-sample recovery guarantees for a tractable sparse-additive variant. Empirically, MOSAIC recovers domain-consistent variable groups across RNA molecular dynamics, solar wind, ENSO climate, the Tennessee Eastman process, and a synthetic tokamak benchmark, enabling interpretable discovery of latent mechanisms in scientific time series.

  • 7 authors
·
May 5

How Much Do Circuits Tell Us? Measuring the Consistency and Specificity of Language Model Circuits

The circuits framework in mechanistic interpretability aims to identify sparse subgraphs of model components that are causally responsible for a behavior, typically evaluated by measuring necessity and sufficiency. But these criteria say little about whether a circuit consistently captures how a model performs a task, or if it is specific to that task. We study these two properties, consistency and specificity, across six tasks and five models, extracting circuits at the component level (attention heads and MLP blocks) and at the level of individual MLP neurons. We find that component-level circuits are highly consistent and causally important for most tasks, but they are not specific: for a given task, ablating its own circuit drops accuracy by about as much as ablating another task's circuit. Neuron-level circuits, on the other hand, exhibit higher specificity across tasks, but far less consistency within tasks. This is explained by circuit overlap: component-level circuits share most of their components across tasks, related or not, while neuron-level circuits overlap only between closely related tasks. In a case study of the components shared by the task circuits of Llama-3.2-3B, we show that they consist mostly of MLP blocks, while the few attention heads within turn out to have general-purpose roles. Overall, our findings raise questions about the degree to which circuits can support targeted understanding of, and intervention on, model behavior.

  • 2 authors
·
Oct 6

DrugTargetWorld: A Synthetic Biobank for Training and Benchmarking AI Scientists

Drug target discovery requires distinguishing molecules that causally drive disease from those that are merely associated with it. Training and evaluating AI agents to perform this workflow end-to-end is difficult because real world biobanks lack known causal ground truth and participant-level data is access controlled. We introduce DrugTargetWorld, a framework that procedurally generates simulated biobanks, or "worlds," with known but concealed causal structure. Each world contains genotypes, proteins, health records, outcomes, and synthetic magnetic resonance imaging (MRI) for 54,000 participants. Agents must construct a disease phenotype, identify causal driver proteins, infer the beneficial direction of modulation, and optionally conduct virtual 'wet lab' experiments. We evaluated nine agents in 540 episodes across 20 cardiovascular worlds and three experimental budgets. Opus 5 and GPT-5.6 Sol achieved the highest mean composite scores, 39.98 and 35.38 of 100, respectively, and both recovered 64% of causal drivers on average. However, no agent reliably distinguished misleading non-causal proteins, and performance remained limited by the integrative judgments required to connect phenotype construction, causal evidence, and intervention decisions. By making each world's causal structure known to the evaluator but hidden from the agent, DrugTargetWorld turns end-to-end drug target discovery into a scalable training and evaluation problem with verifiable reward.

  • 15 authors
·
Oct 6

Learning Transformer Programs

Recent research in mechanistic interpretability has attempted to reverse-engineer Transformer models by carefully inspecting network weights and activations. However, these approaches require considerable manual effort and still fall short of providing complete, faithful descriptions of the underlying algorithms. In this work, we introduce a procedure for training Transformers that are mechanistically interpretable by design. We build on RASP [Weiss et al., 2021], a programming language that can be compiled into Transformer weights. Instead of compiling human-written programs into Transformers, we design a modified Transformer that can be trained using gradient-based optimization and then automatically converted into a discrete, human-readable program. We refer to these models as Transformer Programs. To validate our approach, we learn Transformer Programs for a variety of problems, including an in-context learning task, a suite of algorithmic problems (e.g. sorting, recognizing Dyck languages), and NLP tasks including named entity recognition and text classification. The Transformer Programs can automatically find reasonable solutions, performing on par with standard Transformers of comparable size; and, more importantly, they are easy to interpret. To demonstrate these advantages, we convert Transformers into Python programs and use off-the-shelf code analysis tools to debug model errors and identify the "circuits" used to solve different sub-problems. We hope that Transformer Programs open a new path toward the goal of intrinsically interpretable machine learning.

  • 3 authors
·
Jun 1, 2023 1

Towards Explainable Anticancer Compound Sensitivity Prediction via Multimodal Attention-based Convolutional Encoders

In line with recent advances in neural drug design and sensitivity prediction, we propose a novel architecture for interpretable prediction of anticancer compound sensitivity using a multimodal attention-based convolutional encoder. Our model is based on the three key pillars of drug sensitivity: compounds' structure in the form of a SMILES sequence, gene expression profiles of tumors and prior knowledge on intracellular interactions from protein-protein interaction networks. We demonstrate that our multiscale convolutional attention-based (MCA) encoder significantly outperforms a baseline model trained on Morgan fingerprints, a selection of encoders based on SMILES as well as previously reported state of the art for multimodal drug sensitivity prediction (R2 = 0.86 and RMSE = 0.89). Moreover, the explainability of our approach is demonstrated by a thorough analysis of the attention weights. We show that the attended genes significantly enrich apoptotic processes and that the drug attention is strongly correlated with a standard chemical structure similarity index. Finally, we report a case study of two receptor tyrosine kinase (RTK) inhibitors acting on a leukemia cell line, showcasing the ability of the model to focus on informative genes and submolecular regions of the two compounds. The demonstrated generalizability and the interpretability of our model testify its potential for in-silico prediction of anticancer compound efficacy on unseen cancer cells, positioning it as a valid solution for the development of personalized therapies as well as for the evaluation of candidate compounds in de novo drug design.

  • 6 authors
·
Apr 25, 2019

The Dead Salmons of AI Interpretability

In a striking neuroscience study, the authors placed a dead salmon in an MRI scanner and showed it images of humans in social situations. Astonishingly, standard analyses of the time reported brain regions predictive of social emotions. The explanation, of course, was not supernatural cognition but a cautionary tale about misapplied statistical inference. In AI interpretability, reports of similar ''dead salmon'' artifacts abound: feature attribution, probing, sparse auto-encoding, and even causal analyses can produce plausible-looking explanations for randomly initialized neural networks. In this work, we examine this phenomenon and argue for a pragmatic statistical-causal reframing: explanations of computational systems should be treated as parameters of a (statistical) model, inferred from computational traces. This perspective goes beyond simply measuring statistical variability of explanations due to finite sampling of input data; interpretability methods become statistical estimators, and findings should be tested against explicit and meaningful alternative computational hypotheses, with uncertainty quantified with respect to the postulated statistical model. It also highlights important theoretical issues, such as the identifiability of common interpretability queries, which we argue is critical to understand the field's susceptibility to false discoveries, poor generalizability, and high variance. More broadly, situating interpretability within the standard toolkit of statistical inference opens promising avenues for future work aimed at turning AI interpretability into a pragmatic and rigorous science.

  • 4 authors
·
Dec 21, 2025