Skip to yearly menu bar Skip to main content


Poster
in
Workshop: Challenges in Deployable Generative AI

Circuit Breaking: Removing Model Behaviors with Targeted Ablation

Maximilian Li · Xander Davies · Max Nadeau

Keywords: [ model editing ] [ toxicity ] [ circuits ] [ ablation ] [ Language Modelling ]


Abstract:

Language models often exhibit behaviors that improve performance on a pre-training objective but harm performance on downstream tasks. We propose a novel approach to removing undesirable behaviors by ablating a small number of causal pathways between model components, with the intention of disabling the computational circuit responsible for the bad behavior. Given a small dataset of inputs where the model behaves poorly, we learn to ablate a small number of important causal pathways. In the setting of reducing GPT-2 toxic language generation, we find ablating just 12 of the 11.6K causal edges mitigates toxic generation with minimal degradation of performance on other inputs.

Chat is not available.