hypothesis
active
hypothesis:more-investigator-runs-would-eventually-yield-a-win-with-three-dataset-saeMore investigator runs would eventually yield a win with three-dataset SAE
Cf. 1/35 chat+grep win.
Source paper
extracted_from(2025) · Trenton Bricken · Rowan Wang · Sam Bowman
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- RM features at rank 181, outside top 100.
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.
- Prediction following diversity-undertraining logic.
- Investigator never won using pre-fill attacks and rarely with raw query tool, unlike human teamsfinding0.741Agent fails to use tools like humans.
- Realistic affordances result.
- Claim that feature grounding enables interpretability metrics.
- No features correspond to Oxford study.
- Authors' hedged interpretation.
- Fig 2 right.
- App. B.