method
active
method:pairwise-steering-evaluationPairwise Steering Evaluation
Named procedure for simultaneously injecting two persona vectors and measuring joint trait-expression outcomes
Neighborhood — ranked by edge-count
Papers (1)
paper
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Novel method that applies intervention only when the model begins a new thinking step (at the \n\n delimiter) rather than at every token
- Ability to steer model behavior in two opposite semantic directions on a trait.
- Cosine similarity is useful but insufficient; highly similar traits can still destructively interact
- Method where Kimi K2.5 steers its own SAE features in real time and reports on its internal emotional state
- Paradigm of finding the right direction in activation space (e.g., linear steering).
- The method can steer the model in both positive and negative directions on the target semantic.
- A method for modifying model behavior by adding perturbation vectors to activations, used here to try to reduce eval awareness.
- Classification scheme for pairwise persona vector composition outcomes: constructive, dominant, or destructive