framework
active
framework:two-category-taxonomy-of-sae-transcoder-concerns-faithfulness-vs-simplicityTwo-category taxonomy of SAE/transcoder concerns (faithfulness vs simplicity)
Refactoring of concerns into mechanistic faithfulness and simplicity.
Neighborhood — ranked by edge-count
Papers (1)
paper
- A Toy Model of Mechanistic (Un)Faithfulnessintroducesmentions
Concepts (2)
concept
- Idea that decomposed components must causally reflect the model's computation; enforced by adversarial ablation in VPD.
- Second category of SAE/transcoder concerns, distinct from faithfulness.
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.
- Core motivation.
- Claim that feature grounding enables interpretability metrics.
- Extension of mechanistic interpretability findings to the metacognitive domain
- Surprising finding that the two evaluation methods diverge in their relationship with persistence
- Interpretability method criticized in this paper for shattering manifolds into atomic pieces, obscuring overarching semantic structure.
- SAEs uncover safety-relevant representations that might be monitored or controlled.
- Motivating example with 13 vs 3 features.
- Indicates redundant structure across layers.