Activation Space Interventions Can Be Transferred Between Large Language Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Oozeer, Narmeen, Nathawani, Dhruv, Prakash, Nirmalendu, Lan, Michael, Harrasse, Abir, Abdullah, Amirali
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914046603689984
author Oozeer, Narmeen
Nathawani, Dhruv
Prakash, Nirmalendu
Lan, Michael
Harrasse, Abir
Abdullah, Amirali
author_facet Oozeer, Narmeen
Nathawani, Dhruv
Prakash, Nirmalendu
Lan, Michael
Harrasse, Abir
Abdullah, Amirali
contents The study of representation universality in AI models reveals growing convergence across domains, modalities, and architectures. However, the practical applications of representation universality remain largely unexplored. We bridge this gap by demonstrating that safety interventions can be transferred between models through learned mappings of their shared activation spaces. We demonstrate this approach on two well-established AI safety tasks: backdoor removal and refusal of harmful prompts, showing successful transfer of steering vectors that alter the models' outputs in a predictable way. Additionally, we propose a new task, \textit{corrupted capabilities}, where models are fine-tuned to embed knowledge tied to a backdoor. This tests their ability to separate useful skills from backdoors, reflecting real-world challenges. Extensive experiments across Llama, Qwen and Gemma model families show that our method enables using smaller models to efficiently align larger ones. Furthermore, we demonstrate that autoencoder mappings between base and fine-tuned models can serve as reliable ``lightweight safety switches", allowing dynamic toggling between model behaviors.
format Preprint
id arxiv_https___arxiv_org_abs_2503_04429
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Activation Space Interventions Can Be Transferred Between Large Language Models
Oozeer, Narmeen
Nathawani, Dhruv
Prakash, Nirmalendu
Lan, Michael
Harrasse, Abir
Abdullah, Amirali
Artificial Intelligence
The study of representation universality in AI models reveals growing convergence across domains, modalities, and architectures. However, the practical applications of representation universality remain largely unexplored. We bridge this gap by demonstrating that safety interventions can be transferred between models through learned mappings of their shared activation spaces. We demonstrate this approach on two well-established AI safety tasks: backdoor removal and refusal of harmful prompts, showing successful transfer of steering vectors that alter the models' outputs in a predictable way. Additionally, we propose a new task, \textit{corrupted capabilities}, where models are fine-tuned to embed knowledge tied to a backdoor. This tests their ability to separate useful skills from backdoors, reflecting real-world challenges. Extensive experiments across Llama, Qwen and Gemma model families show that our method enables using smaller models to efficiently align larger ones. Furthermore, we demonstrate that autoencoder mappings between base and fine-tuned models can serve as reliable ``lightweight safety switches", allowing dynamic toggling between model behaviors.
title Activation Space Interventions Can Be Transferred Between Large Language Models
topic Artificial Intelligence
url https://arxiv.org/abs/2503.04429