Papers/2611.17481
🧪 Test?View on arXiv

Creating Steering Vectors via Input Optimisation

Not provided

input optimizationlanguage model steeringactivation manipulation
2611.17481
Builder Relevance
70%
19h ago

Abstract

The paper introduces ActGCG, a method for optimizing input to influence language model behavior without requiring access to model internals.

Reality Card

Core Claim

ActGCG successfully optimizes input to match target activations, allowing steering of language models without internal access.

Method / Result

A discrete suffix reproduces the steering vector's behavior almost perfectly across prompts and models.

Limitations

The equivalence for anti-refusal steering is weaker, succeeding only partially under strict evaluation metrics.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.
← Back to all papers