Fazl Barez

Beyond Linear Probes: Dynamic Safety Monitoring for Language Models

Towards Understanding Subliminal Learning: When and How Hidden Biases Transfer

Query Circuits: Explaining How Language Models Answer User Prompts

Measuring Sparse Autoencoder Feature Space Similarities Across Large Language Models

Understanding Addition and Subtraction in Transformers

VAL-Bench: Measuring value alignment in Language Models

Chain-of-Thought Hijacking

Towards Interpreting Visual Information Processing in Vision-Language Models

Best-of-N Jailbreaking

Rethinking Safety in LLM Fine-tuning: An Optimization Perspective

Sparse Autoencoders Reveal Universal Feature Spaces Across Large Language Models

Scaling Sparse Feature Circuits For Studying In-Context Learning

PoisonBench: Assessing Large Language Model Vulnerability to Data Poisoning

PoisonBench: Assessing Language Model Vulnerability to Poisoned Preference Data

Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoders

Plan B: Training LLMs to fail less severely

Attacking Audio Language Models with Best-of-N Jailbreaking