claim
active
claim:contrastive-scoring-weakly-penalises-polysemanticity-making-it-a-useful-measure-for-comparing-saes-and-autointerp-methodsContrastive scoring weakly penalises polysemanticity, making it a useful measure for comparing SAEs and autointerp methods
Justification of scoring.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Shown in graph for the nearest-neighbour test.
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.
- Better bases may give sparsity.
- Major prediction checked against InceptionV1.
- Main proposal.
- Probing limited.
- Lower SAE loss at fixed sparsity likely indicates less interference and less superposition.claim0.742Justifies SAE proxy.
- Motivates multi-metric evaluation.
- Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.
- If agentic self-steering evaluation proves robust, it might be used to better explain and interpret SAE features in generalhypothesis0.736Speculative claim about scaling introspective access to general SAE feature interpretation