finding
active
finding:residual-stream-probing-gives-94-accuracy-on-trained-pythia-1b-vs-84-on-random-modelResidual-stream probing gives 94% accuracy on trained Pythia-1B vs 84% on random model
Random baseline context.
Source paper
extracted_from(2025) · Adam Karvonen · Can Rager · Johnny Lin · Curt Tigges +11
Neighborhood — ranked by edge-count
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.
- Supervised metrics discriminate.
- SAEs on random Pythia-1B achieve higher TPP than trained-model SAEs, attributed to lower probe baselinefinding0.826Caveated result.
- Attributed to model anisotropy from saturation making hidden states harder to access
- Linear readout works but is not the true representation.
- Case Study II result showing DAS identifies fewer causally relevant positions than a probe
- Linear probe achieves 100% classification accuracy for almost all components in Pythia-6.9B gender taskfinding0.778Demonstrates that linear probes can overestimate causal relevance; probes succeed on non-causally-relevant representations
- Baseline accuracy showing small models fail on harder NPI licensing tasks
- Justifies restricting probe-based vector derivation to h_b activations; attributed to Yes/No semantics