claim
active
claim:privileged-access-effects-may-be-fragile-or-need-more-training-data-before-emerging-in-broad-distributionsPrivileged-access effects may be fragile or need more training data before emerging in broad distributions
Possible explanations for difference from Li et al.
Source paper
extracted_from(2026) · Adam Karvonen · Euan Ong · Subhash Kantamneni · Samuel Marks
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- No evidence of privileged access.
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.
- Assumption D.3 formalized in the theoretical framework; empirically validated with coin-flip sequence experiments
- Information about own internal state available only to the model.
- Motivation.
- If the relationship is present in activations, better tools (e.g. trained on these datasets) could surface ithypothesis0.737Alternative explanation.
- SAEs uncover safety-relevant representations that might be monitored or controlled.
- Models predict their own hypothetical behavior better than other models can, demonstrating a form of privileged self-access per Binder et al. 2024
- Core framing.
- Claim that capability emerges from architecture, not data, and that later models lose the surprise.