claim
active
claim:the-paper-makes-no-claims-about-the-provenance-of-model-exclusive-features-training-vs-data-artifactsThe paper makes no claims about the provenance of model-exclusive features (training vs data artifacts).
Limitation.
Source paper
extracted_from(2026) · Thomas Jiralerspong · Trenton Bricken
Neighborhood — ranked by edge-count
Papers (1)
paper
Questions (1)
question
- Future work.
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.
- Primary unexpected empirical observation.
- Authors argue features are model properties because logit effects and ablations are consistent with feature interpretations
- Appendix C.5/C.7.
- Primary limitation acknowledged by the authors; strongest evidence would require mechanistic activation analysis
- Authors take agnostic position on ontological status but universality evidence pushes toward features being real
- Key limitation identified: NLAs hallucinate specific details while preserving thematic accuracy; informs practical usage.
- Summary assertion that traditional evidence fails for novel agents.
- Toy model in the abundant-capacity regime.