9 papers · 1 filter
Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration
Aashiq Muhamed, Mona T. Diab, Virginia Smith
Safety guardrails in open-weight language models can be readily bypassed using Refusal Feature Ablation (RFA), a technique that identifies and projects out a linear refusal directi…
Pick Your Poison: Learning to Select Poison Sets for Stronger LLM Backdoor Attacks
Aashiq Muhamed, Mona T. Diab, Virginia Smith +2
Backdoor poisoning attacks add poisoned examples to otherwise-clean finetuning data, pairing a trigger with a target behavior that the model learns to produce when the trigger appe…
MOLE: Detecting Insider Threats in AI Agents
Aashiq Muhamed, Virginia Smith
Model misalignment, prompt injection, or operator misuse could lead AI agents operating frontier-lab accounts to exfiltrate model weights, poison training data, or weaken release g…
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.…
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 Can 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…
Decoding Dark Matter: Specialized Sparse Autoencoders for Interpreting Rare Concepts in Foundation Models
Aashiq Muhamed, Mona Diab, Virginia Smith
Understanding and mitigating the potential risks associated with foundation models (FMs) hinges on developing effective interpretability methods. Sparse Autoencoders (SAEs) have em…
Grass: Compute Efficient Low-Memory LLM Training with Structured Sparse Gradients
Aashiq Muhamed, Oscar Li, David Woodruff +2
Large language model (LLM) training and finetuning are often bottlenecked by limited GPU memory. While existing projection-based optimization methods address this by projecting gra…