finding
active
finding:flipped-label-sentiment-task-with-three-labels-predicted-at-cross-validated-r-0-96-needing-substantially-more-shotsFlipped-label sentiment task with three labels predicted at cross-validated r=0.96, needing substantially more shots
Non-persona generality.
Source paper
extracted_from(2025) · Eric Bigelow · Daniel Wurgaft · YingQiao Wang · Noah D. Goodman +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Core 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.
- Held-out prediction protocol.
- Finding that the two evaluation modalities frequently diverge in their interpretation of the same SAE feature
- Claim that S predicts threshold midpoints across different bases, tasks, and models
- Five judge models agree 90-96% on multi-attempt detection and ESR direction for same responsesfinding0.732Validation that ESR findings are not artifacts of any particular judge model's evaluation methodology
- Demonstration of strong prior rebinding via small coherent anchors
- Steering effect size across 35 emotion vectors correlates with probe-preference correlation at r=0.85finding0.727Emotions correlating with preference causally drive it.
- Predicts phase boundaries.
- Negative probe predicts dispreference.