Search papers, labs, and topics across Lattice.
This paper introduces Prompt Steering Replacement (PSR), a method for activation steering that mimics the token-specific interventions of prompt steering. They find that existing activation steering methods are not faithful to prompt steering, which applies strong interventions on some tokens while barely affecting others. PSR models are trained to estimate token-specific steering coefficients from activations and outperform existing activation steering methods on three steering benchmarks, even rivaling prompting in some cases.
Activation steering can finally match the nuanced control of prompt engineering: token-specific interventions learned from prompts let you steer LLMs more effectively.
Large language models can be steered at inference time through prompting or activation interventions, but activation steering methods often underperform compared to prompt-based approaches. We propose a framework that formulates prompt steering as a form of activation steering and investigates whether distilling successful prompt steering behavior into simpler, interpretable models can close this gap. Our analysis reveals that popular activation steering methods are not faithful to the mechanics of prompt steering, which applies strong interventions on some tokens while barely affecting others. Based on these insights, we introduce Prompt Steering Replacement (PSR) models that estimate token-specific steering coefficients from the activations themselves and are trained to imitate prompt-based interventions. Experiments on three steering benchmarks across multiple language models show that PSR models outperform existing activation steering methods, especially when controlling for high-coherence completions, and also compare favorably to prompting on AxBench and persona steering.