Steering Language Models with Activation Engineering

ICLR 2025 Conference Submission7934 Authors

Keywords: interpretability, steering, alignment, safety, sentiment

TL;DR: We steer LLMs by adding a bias term computed from the model's representations of simple prompt pairs.

Abstract: Prompt engineering and finetuning aim to maximize language model performance on a given metric (like toxicity reduction). However, these methods do not optimally elicit a model's capabilities. To reduce this gap, we introduce a form of _activation engineering_: the inference-time modification of activations in order to control (or _steer_) model outputs. Specifically, we introduce the Activation Addition (ActAdd) technique, which contrasts the intermediate activations on prompt pairs (such as “Love” versus “Hate”) to compute a _steering vector_. By tactically adding in e.g. the “Love”$-$“Hate” steering vector during the forward pass, ActAdd can perform many tasks like topic steering, sentiment steering, and detoxification. ActAdd yields inference-time control over high-level output properties (like topic and sentiment) while preserving performance on off-target tasks. ActAdd is lightweight: it does not require any machine optimization and works with a single pair of data points, which enables rapid iteration over steering.

Primary Area: alignment, fairness, safety, privacy, and societal considerations

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7934