#artificial intelligence
Jul 2026
Where Steering Signals Come From: Activation Source Selection in Activation Steering
Tail subtraction is introduced, which removes shared prompt and continuation semantics from boundary states and yields cleaner, more stable steering signals, and suggests that steering depends on representations of what the model is about to do, not merely on what has already appeared.
Jiaran Ye, Lingxu Ran, Zijun Yao et al.
· arXiv.org · 2 citations