Evan Hubinger — accepted AI-conference papers

2024

Steering Llama 2 via Contrastive Activation Addition

Nina Rimsky, Nick Gabrieli, Julian Schulz, Meg Tong, Evan Hubinger, Alexander Turner

ACL 2024long

We introduce Contrastive Activation Addition (CAA), a method for steering language models by modifying their activations during forward passes. CAA computes “steering vectors” by averaging the difference in residual stream activations between pairs of positive and negative examples of a particular b…

Cited by 132 Source ↗PDF ↗code ↗Scholar ↗

2023

Discovering Language Model Behaviors with Model-Written Evaluations

Ethan Perez, Sam Ringer, Kamile Lukosiute, Karina Nguyen, Edwin Chen, Scott Heiner +57

ACL 2023findings

As language models (LMs) scale, they develop many novel behaviors, good and bad, exacerbating the need to evaluate how they behave. Prior work creates evaluations with crowdwork (which is time-consuming and expensive) or existing data sources (which are not always available). Here, we automatically…

Cited by 290 Source ↗PDF ↗code ↗Scholar ↗