Aidan Ewart

Mechanistic Unlearning: Robust Knowledge Unlearning and Editing via Mechanistic Localization

Mechanistic Unlearning: Robust Knowledge Unlearning and Editing via Mechanistic Localization

Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMs

Sparse Autoencoders Find Highly Interpretable Features in Language Models