finding
active
finding:saes-on-random-pythia-1b-achieve-higher-tpp-than-trained-model-saes-attributed-to-lower-probe-baselineSAEs on random Pythia-1B achieve higher TPP than trained-model SAEs, attributed to lower probe baseline
Caveated result.
Source paper
extracted_from(2025) · Adam Karvonen · Can Rager · Johnny Lin · Curt Tigges +11
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Supervised metrics discriminate.
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.
- Random baseline context.
- Case Study II result showing DAS identifies fewer causally relevant positions than a probe
- Evidence in real models.
- Attributed to model anisotropy from saturation making hidden states harder to access
- Surprising negative result for LDA despite being a supervised method
- Probing limited.
- Baseline accuracy showing small models fail on harder NPI licensing tasks
- Smaller models' weaker concept representations explain why supervised metrics differ on Pythia-160Mclaim0.746Interpretation of model-scale effects.