claim
active
claim:arbitrarily-expressive-nonlinear-activation-functions-raise-concerns-about-whether-a-feature-is-truly-representedArbitrarily expressive nonlinear activation functions raise concerns about whether a feature is truly represented
Contrast with simple gate MLPs.
Source paper
extracted_from(2025) · Lucius Bushnaq · Dan Braun · Lee Sharkey
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.
- Open mechanistic question arising from the causal steering experiment
- Limitation acknowledgment about the adequacy of the linear representation assumption
- Explains how non-separability shifts identity from parts to collective.
- Motivates shift from studying model activations ('thoughts') to understanding parameters ('the computations themselves').
- Shows interpretability correlates with activation strength, most model effect comes from high activations
- Mechanism by which activation of an emotion feature sometimes leads to later suppression of that same featurequestion0.742Identified research gap: the paper observes anti-persistence but has no explanation for it
- Steering derived from LRH.
- Motivates nested SAE.