Papers with Mechanistic interpretability
Emergence of Minimal Circuits for Indirect Object Identification in Attention-Only Transformers (2026.acl-srw)
Copied to clipboard
| Challenge: | Large language models are difficult to reverse engineer because of their internal operations. |
| Approach: | They train small attention-only transformers on a symbolic version of the Indirect Object Identification task. |
| Outcome: | The proposed model with only two attention heads achieves perfect IOI accuracy despite lacking MLPs and normalization layers . |
Mechanistic Analysis Of Universality: Numerical Comparison Circuits Across Transformer Architectures (2026.acl-srw)
Copied to clipboard
| Challenge: | Mechanistic interpretability seeks to identify internal circuits within transformer language models but it is unclear whether they generalize across model families and scales. |
| Approach: | They propose to identify internal circuits within transformer language models by numerical comparisons. |
| Outcome: | The proposed model implementations are consistent across architecture and scale, the authors show . their results highlight the need for cross model comparisons to claim generalization of internal circuits. |
Route Sparse Autoencoder to Interpret Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Sparse autoencoders (SAEs) extract interpretable and monosemantic features in large language models . prior work focused on feature extraction from a single layer, failing to capture activations that span multiple layers. |
| Approach: | They propose a framework that integrates a routing mechanism with a shared SAE to efficiently extract features from multiple layers. |
| Outcome: | The proposed framework extracts features from multiple layers while incurring minimal parameter overhead while achieving high interpretability and flexibility. |
A Simple Yet Effective Method for Non-Refusing Context Relevant Fine-grained Safety Steering in LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning large language models to meet safety policies are costly and impractical. |
| Approach: | They propose a method to fine-tune large language models to meet evolving safety policies by applying a gradient-free, unsupervised approach. |
| Outcome: | The proposed method provides precise control, avoids blanket refusals, and directs models to generate safe, relevant content. |