arXiv:2601. 21996v2 Announce Type: replace-cross Abstract: While Mechanistic Interpretability has identified interpretable circuits in LLMs, their causal origins in training data remain elusive.
arXiv:2606. 29171v1 Announce Type: cross Abstract: While existing data attribution methods can identify which training examples build specific mechanistic circuits, they cannot explain how training data shapes the high-level behavioral decisions a model learns to make.
The paper investigates how LLaMA 3.1‑8B models numerical sequence patterns, focusing on time‑series prediction. By designing a task that requires detecting structural cues—specifically first differences in a sequence—the authors show that the model performs well and internally computes and stores these differences. Probing and activation‑patching experiments reveal that LLaMA retrieves and applies the first‑difference via an induction‑like circuit, marking one of the first demonstrations of concept induction in large language models.
arXiv:2605. 28854v2 Announce Type: replace-cross Abstract: Large language models (LLMs) exhibit remarkable flexibility in adapting to novel tasks from in-context examples without parameter updates, a capability known as in-context learning (ICL).
arXiv:2608. 12036v1 Announce Type: new Abstract: AI models have achieved remarkable success across diverse domains, yet the mechanisms underlying their capabilities and the risks they may pose remain poorly understood.
arXiv:2507. 06445v2 Announce Type: replace-cross Abstract: Interpretability research often predicts model responses to targeted mechanistic interventions.
arXiv:2607. 07316v1 Announce Type: new Abstract: This article offers a comprehensive overview of mechanistic interpretability, an emerging field that seeks to reverse-engineer the internal algorithms of modern neural networks.
Matryoshka Attribution (MAttr) is a mask‑learning method that identifies nested subsets of a language model’s internal components by minimizing downstream loss. It uses a differentiable sigmoid top‑k operator and randomizes sparsity during training to produce an attribution ordering of components. MAttr tops the Mechanistic Interpretability Benchmark leaderboard and can be applied via reinforcement learning to pinpoint weight changes that control behaviors such as refusal in Llama 3.1 8B Instruct, where restoring just 1% of weights removes refusals while preserving capabilities.
The paper introduces a circuit‑grounded framework that links training‑dynamics‑based data valuation with mechanistic interpretability. It defines data quality along learnability, challenge, and alignment, identifies internal model circuits that control these utilities, and uses them as controllable interfaces for data generation. The authors present SAMS, a stage‑aware scheduling method that steers circuit‑guided data to match the model’s evolving optimization needs, achieving more diverse and effective data than prompt‑based baselines on multiple‑choice QA tasks.
arXiv:2608.24482v1 Announce Type: cross Abstract: Mechanistic Localization bridges mechanistic interpretability and post-training optimization by isolating critical parameters via interpretative appr...
arXiv:2609.15064v1 Announce Type: new Abstract: Reinforcement learning (RL) is widely utilized in large language model training to improve targeted capabilities, yet how RL reshapes a model remains p...
arXiv:2510. 12957v4 Announce Type: replace-cross Abstract: We treat the internals of generative models as mechanistic objects rather than black boxes.
The paper introduces Concept-Targeted Attribution (CTA), a method that trains attribution graphs to explain the emergence of internal concept representations in language models, rather than just the final token prediction. CTA produces probe-specific circuits that reveal which internal computations drive a linear probe’s accuracy, and cross-layer transcoders demonstrate that these graphs contain predictive structure across multiple concept categories. Causal ablations show that probe-targeted and logit-targeted graphs capture distinct mechanisms, with probe-relevant features affecting internal concept scores and logit-relevant features altering generated tokens.