TechPrufer
Get started
Apple
Apple logoApple · Feature trail

Dynamically Scaled Activation Steering

2 posts · Sep 2026

What this trail covers

This post investigates the effectiveness-fluency trade-off in LLM conditioning methods. It finds that efficient steering methods often degrade fluency, and that activation steering is less effective on instruction-tuned models compared to base models. Prompting and fine-tuning are better for concept injection but not removal. Textual metrics correlate well with LLM-as-judge scores and provide insights into conditioning behavior.

All Apple trails
  1. On the Effectiveness-Fluency Trade-Off in LLM Conditioning: A Systematic StudyThis post systematically studies the effectiveness-fluency trade-off in LLM conditioning methods, identifying that efficient steering methods often come at a cost to fluency. It also highlights a critical interaction with training paradigms, showing that activation steering is less effective on instruction-tuned models. The study finds that simple prompting and supervised fine-tuning are viable for concept injection but not removal, and that textual metrics correlate with LLM-as-judge scores.
  2. Dynamically Scaled Activation SteeringIntroduced Dynamically Scaled Activation Steering (DSAS), a method-agnostic framework that decouples when to steer from how to steer. DSAS adaptively modulates the strength of existing steering transformations across layers and inputs, intervening strongly only when undesired behavior is detected. At generation time, DSAS computes context-dependent scaling factors that selectively adjust the strength of any steering method. Demonstrated DSAS's generality by applying it to a text-to-image diffusion model, showing how adaptive steering allows the modulation of specific concepts. DSAS introduces minimal computational overhead while improving interpretability, pinpointing which tokens require steering and by how much.