claim
active
claim:saes-do-not-solve-the-overarching-problem-of-mechanistic-interpretability-but-are-a-central-component-of-today-s-toolboxSAEs do not solve the overarching problem of mechanistic interpretability but are a central component of today's toolbox
Authors' hedged position on SAEs.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Under the Hood of a Reasoning Model - Goodfirecitesmentions
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.
- Explains weaker datamix SAEs.
- Broadly, turn-averaged SAEs make interpretability techniques practical at long context lengths.quote0.810Abstract thesis.
- Extension of mechanistic interpretability findings to the metacognitive domain
- Automated interpretability and specificity ratings show SAE features are clearer than MLP neurons.
- Claim that feature grounding enables interpretability metrics.
- Core critique of sparse autoencoders: they break the geometric structure of representations, making it harder to see the big picture.
- Geometric explanation for prior negative results.