Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ali, Sheikh Abdur Raheem, Xu, Justin, Yang, Ivory, Li, Jasmine Xinze, Arslan, Ayse, Benham, Clark
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2507.11771
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911059129925632
author Ali, Sheikh Abdur Raheem
Xu, Justin
Yang, Ivory
Li, Jasmine Xinze
Arslan, Ayse
Benham, Clark
author_facet Ali, Sheikh Abdur Raheem
Xu, Justin
Yang, Ivory
Li, Jasmine Xinze
Arslan, Ayse
Benham, Clark
contents As large language models (LLMs) evolve in complexity and capability, the efficacy of less widely deployed alignment techniques are uncertain. Building on previous work on activation steering and contrastive activation addition (CAA), this paper explores the effectiveness of CAA with model scale using the family of Llama 2 models (7B, 13B, and 70B). CAA works by finding desirable 'directions' in the model's residual stream vector space using contrastive pairs (for example, hate to love) and adding this direction to the residual stream during the forward pass. It directly manipulates the residual stream and aims to extract features from language models to better control their outputs. Using answer matching questions centered around the refusal behavior, we found that 1) CAA is most effective when applied at early-mid layers. 2) The effectiveness of CAA diminishes with model size. 3) Negative steering has more pronounced effects than positive steering across all model sizes.
format Preprint
id arxiv_https___arxiv_org_abs_2507_11771
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Scaling laws for activation steering with Llama 2 models and refusal mechanisms
Ali, Sheikh Abdur Raheem
Xu, Justin
Yang, Ivory
Li, Jasmine Xinze
Arslan, Ayse
Benham, Clark
Machine Learning
As large language models (LLMs) evolve in complexity and capability, the efficacy of less widely deployed alignment techniques are uncertain. Building on previous work on activation steering and contrastive activation addition (CAA), this paper explores the effectiveness of CAA with model scale using the family of Llama 2 models (7B, 13B, and 70B). CAA works by finding desirable 'directions' in the model's residual stream vector space using contrastive pairs (for example, hate to love) and adding this direction to the residual stream during the forward pass. It directly manipulates the residual stream and aims to extract features from language models to better control their outputs. Using answer matching questions centered around the refusal behavior, we found that 1) CAA is most effective when applied at early-mid layers. 2) The effectiveness of CAA diminishes with model size. 3) Negative steering has more pronounced effects than positive steering across all model sizes.
title Scaling laws for activation steering with Llama 2 models and refusal mechanisms
topic Machine Learning
url https://arxiv.org/abs/2507.11771