concept
active
concept:uncertainty-aware-dynamic-steeringUncertainty-aware dynamic steering
Proposed future direction: model dynamically adjusts steering strength based on internal uncertainty during inference
Neighborhood — ranked by edge-count
Claims (1)
claim
- Interpretive claim from probing experiment showing reflection direction features outperform baseline for uncertainty prediction
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.
- Forward-looking claim connecting uncertainty-reflection hypothesis to practical future work
- Steering variants that adaptively modulate steering strength based on input context or token position, addressing when to steer
- Current steering applies fixed strength; dynamic uncertainty-aware steering during inference is an open gapquestion0.791Research gap identified in limitations/future work section connecting uncertainty findings to practical improvement
- Subtracting a scaled persona vector from hidden states at each decoding step to reduce trait expression after finetuning
- Replicates main result using in-distribution steering vector; addresses concern about pre-trained vector validity.
- Framework of using internal-state representations to control or steer generative models; conceptually parallel to manifold steering in language models.
- Paradigm of finding the right direction in activation space (e.g., linear steering).
- The central phenomenon introduced by this paper: inference-time recovery from irrelevant activation steering in LLMs