claim
active
claim:roman-numeral-activation-indicates-the-model-learned-semantic-date-representations-beyond-character-matchingRoman numeral activation indicates the model learned semantic date representations beyond character matching
Interpretation of Feature 425104.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Greater-than feature.
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.
- Post-trained models repurpose user token activations to represent the Assistant's thoughts.hypothesis0.733Tested by J-lens comparison.
- Authors argue features are model properties because logit effects and ablations are consistent with feature interpretations
- If the relationship is present in activations, better tools (e.g. trained on these datasets) could surface ithypothesis0.731Alternative explanation.
- Interpretation.
- Universality of Hebrew script feature across two transformers
- Linear representation hypothesis supported by empirical and theoretical arguments.
- Specific tuning curve structure.
- Core goal statement.