activation steering

Definition

Nudging a model's internal state at inference time to push its behavior in a particular direction.

Adding a scaled vector to the residual stream during inference to bias the model toward a target behavior without changing weights.

Also called: steering

Mentioned in 9 episodes

1. 073
   When Three LLMs Talk to Each Other, Their Ideas Quietly Stop Moving
2. 055
   Why LLM Judges Flip Their Verdicts When You Change the Question Format
3. 044
   How One Sentence and a Forged History Flip the Most Aligned Models
4. 040
   Two Frozen Models Learn to Whisper: Coupling Through Hidden States
5. 037
   Why Hallucination Detectors Miss Stale Facts: A Geometric Story About What Models Know But Don't Say
6. 026
   What RL Actually Does to Language Models, at the Token Level
7. 025
   The Missing Gradient Term That Predicts Sycophancy in RLHF
8. 018
   Language Models Compute the Rational Move, Then Override It
9. 006
   What Happens Inside Claude When It Decides to Blackmail Someone

Related concepts