7 papers
Pando: Do Interpretability Methods Work When Models Won't Explain Themselves?
Ziqian Zhong, Aashiq Muhamed, Mona T. Diab +2
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 co…
DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment
James Wedgwood, Aashiq Muhamed, Mona T. Diab +1
Preference alignment is usually achieved by weight-updating training on preference data, which adds substantial alignment-stage compute and provides limited mechanistic visibility.…
RefusalBench: Generative Evaluation of Selective Refusal in Grounded Language Models
Aashiq Muhamed, Leonardo F. R. Ribeiro, Markus Dreyer +2
The ability of language models in RAG systems to selectively refuse to answer based on flawed context is critical for safety, yet remains a significant failure point. Our large-sca…
Position: Mechanistic Interpretability Should Prioritize Feature Consistency in SAEs
Xiangchen Song, Aashiq Muhamed, Yujia Zheng +5
Sparse Autoencoders (SAEs) are a prominent tool in mechanistic interpretability (MI) for decomposing neural network activations into interpretable features. However, the aspiration…
SAEs Improve Unlearning: Dynamic Sparse Autoencoder Guardrails for Precision Unlearning in LLMs
Aashiq Muhamed, Jacopo Bonato, Mona Diab +1
Machine unlearning is a promising approach to improve LLM safety by removing unwanted knowledge from the model. However, prevailing gradient-based unlearning methods suffer from is…
CoRAG: Collaborative Retrieval-Augmented Generation
Aashiq Muhamed, Mona Diab, Virginia Smith
Retrieval-Augmented Generation (RAG) models excel in knowledge-intensive tasks, especially under few-shot learning constraints. We introduce CoRAG, a framework extending RAG to col…