REVIEW 13 cited by
Have Faith in Faithfulness: Going Beyond Circuit Overlap When Finding Model Mechanisms
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Many recent language model (LM) interpretability studies have adopted the circuits framework, which aims to find the minimal computational subgraph, or circuit, that explains LM behavior on a given task. Most studies determine which edges belong in a LM's circuit by performing causal interventions on each edge independently, but this scales poorly with model size. Edge attribution patching (EAP), gradient-based approximation to interventions, has emerged as a scalable but imperfect solution to this problem. In this paper, we introduce a new method - EAP with integrated gradients (EAP-IG) - that aims to better maintain a core property of circuits: faithfulness. A circuit is faithful if all model edges outside the circuit can be ablated without changing the model's performance on the task; faithfulness is what justifies studying circuits, rather than the full model. Our experiments demonstrate that circuits found using EAP are less faithful than those found using EAP-IG, even though both have high node overlap with circuits found previously using causal interventions. We conclude more generally that when using circuits to compare the mechanisms models use to solve tasks, faithfulness, not overlap, is what should be measured.
Forward citations
Cited by 13 Pith papers
-
CURE:Circuit-Aware Unlearning for LLM-based Recommendation
CURE disentangles LLM recommendation circuits into forget-specific, retain-specific, and task-shared modules with tailored update rules to achieve more effective unlearning than weighted baselines.
-
Temporal Preference Concepts and their Functions in a Large Language Model
Temporal preference in Qwen3-4B-Instruct-2507 localizes to layers 17–35 (especially L24 attention), has curved residual-stream geometry, is behaviorally unstable, and can be bidirectionally steered.
-
Validating Causal Abstraction Metrics on Simulated Complex Systems
Authors create a benchmark across discrete/continuous and static/dynamical systems and introduce the Causal Abstraction Error (CAE) metric that reliably distinguishes valid from invalid causal abstractions when it inc...
-
Quantifying the Agreement Between Data-Influence and Data-Similarity to Understand LLM Behavior
Data-similarity and data-influence produce significantly overlapping rankings of training documents for LLM outputs, with asymmetry allowing a favorable cost-accuracy trade-off.
-
Seeing Through Circuits: Faithful Mechanistic Interpretability for Vision Transformers
Edge-based circuits in vision transformers can be automatically recovered to explain and steer model computations for classification and adversarial behaviors.
-
Inside-Out: Measuring Generalization in Vision Transformers Through Inner Workings
Circuit-based metrics from Vision Transformer internals provide better label-free proxies for generalization under distribution shift than existing methods like model confidence.
-
Wired for Overconfidence: A Mechanistic Perspective on Inflated Verbalized Confidence in LLMs
Inflated verbalized confidence in Qwen2.5-3B and Llama-3.2-3B is driven by a compact, cross-dataset set of middle-to-late-layer MLP blocks and attention heads, and steering or ablating those components at inference ti...
-
A Mechanistic Perspective and Circuit-Guided Difficulty Metric for Unlearning
A circuit-similarity score predicts which samples an LLM unlearning method will fail to erase, with hard samples relying on deeper, output-facing pathways.
-
Efficient LLM Adversarial Training via Low-Rank Defense and Circuit-Guided Surrogates
Combining suffix-window representation finetuning with an ActGrad-pruned surrogate cuts latent-adversarial-training FLOPs per step by 48.1% with only 0.0118% trainable parameters, while accepting higher attack success rates.
-
Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces
Reasoning in large output spaces proceeds via shortlisting then fine-grained reasoning; this characterization enables a mechanistic distillation strategy that outperforms standard distillation.
-
Temporal Preference Concepts and their Functions in a Large Language Model
Causal localization via attribution and patching identifies a temporal preference subgraph in mid-to-upper layers of Qwen3-4B-Instruct-2507, with time-horizon geometry in the residual stream and initial evidence for s...
-
$C$-$\Delta\Theta$: Circuit-Restricted Weight Arithmetic for Selective Refusal
Selective refusal can be improved by editing only the small weight circuit found by EAP-IG, yielding offline checkpoints with low over-refusal.
-
A Numerical PDEs Approach to Evolution Equations in Shape Analysis Based on Regularized Morphoelasticity
Regularized morphoelasticity yields a high-order elliptic system for continuous shape evolution that is solved by mixed finite elements in FEniCSx within an LDDMM-style optimal-control growth model.
Discussion (0). Sign in to comment.