claim
active
claim:the-j-lens-is-a-principled-refinement-of-the-logit-lens-correcting-for-representational-change-across-layersThe J-lens is a principled refinement of the logit lens correcting for representational change across layers.
Methods comparison.
Source paper
extracted_fromAuthors
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (2)
finding
- Methodological comparison.
- Tuned lens skips to output answers and is best next-token predictor while J-lens has highest KL.supportsComparison.
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.
- Related work.
- Model organism results.
- Limitation.
- Mediation control.
- Imagine instruction raises property name in J-lens while J-orthogonalized property probe stays at baseline.finding0.790J-space privileged.
- Extension.
- No-instruction baseline rate of target in J-lens is approximately zero across task families.finding0.779Modulation control.
- Implicit modulation.