finding
active
finding:logistic-probes-after-layer-1-achieve-rmse-of-5-for-character-count-with-6-pcs-capturing-82-of-probe-weight-varianceLogistic probes after layer 1 achieve RMSE of 5 for character count, with 6 PCs capturing 82% of probe weight variance
Intrinsic noise in representation.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Central thesis.
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.
- Beliefs linearly decodable.
- Top 6 PCs of 150 mean layer-2 residual vectors capture 95% of variance in the character count subspacefinding0.813Low-dimensional manifold.
- Linear readout works but is not the true representation.
- Probing limited.
- Concrete numerical example showing detection and control are nearly identical at peak apparent accuracy
- Likely-trained MM probe is a surprisingly effective causal baseline due to correlation between truth and probability on sp_en_trans
- Exception to ReLU underperformance.
- Logistic regression trained on GSM8k training set to predict answer correctness from projection features along reflection direction