Concept Steering

Latent intervention technique that manipulates sparse features to steer model predictions toward desired concepts.

Neighborhood — ranked by edge-count

concept

wrecking-ball intervention
associated_with
Type of concept steering intervention that catastrophically collapses global model performance.
Three Operational Regimes of Steering
introduces
The three categories of SAE feature behavior under concept steering identified in the paper

method

Same-concept steering
related_to
Steering using the same concept direction as is being measured, testing whether internal-state shifts causally affect the model's report of that state
Cross-concept steering
related_to
Steering one concept direction while measuring introspection for a different concept, yielding a 4×4 steering-concept × measured-concept matrix to test concept-specific modulability
Target vs. Off-Target Probe Area Metric
implements
Metric introduced to quantify steering selectivity by comparing the area of target and off-target concept probes.

event

Mechanistic Interpretability of EEG Foundation Models via Sparse Autoencoders (2026)
mentions
Preprint applying TopK SAEs to three EEG transformers to reveal sparse feature dictionaries, steering regimes, and spectral interpretation.

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

direction-based steeringconcept0.796
Paradigm of finding the right direction in activation space (e.g., linear steering).
Model Steeringconcept0.790
Using interventions to guide model generation behavior, e.g., adding sentiment vectors at inference time
Representation Steeringconcept0.790
Parent concept; the practice of controlling neural network outputs by manipulating internal representations.
Conceptconcept0.789
Central entity of Jackson's framework: a structure invented to give coherent account of immediate consequences of actions; the building block of software design
steering vectorsconcept0.782
A method for modifying model behavior by adding perturbation vectors to activations, used here to try to reduce eval awareness.
Stepwise steeringmethod0.776
Novel method that applies intervention only when the model begins a new thinking step (at the \n\n delimiter) rather than at every token
geometry-based steeringconcept0.768
Paradigm of finding the right geometry (manifold) for principled control.
concept geometryconcept0.767
The spatial/geometric organization of conceptual structure within neural network representations; central to the paper's thesis.