finding
active
finding:of-5m-entities-65-5-supported-22-9-not-supported-11-6-insufficient-informationOf ~5M entities, 65.5% Supported, 22.9% Not Supported, 11.6% Insufficient Information
Dataset statistics.
Source paper
extracted_from(2026) · Aaditya Vikram Prasad · Connor Watts · Jack Merullo · Dhruvil Gala +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Datasets (1)
dataset
- Entities labeled Supported/Not Supported/Insufficient from 76,849 completions.
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.
- Intervention action distribution.
- Intervals miscalibrated.
- Sampled helpfulness mass.
- Concurrent work result showing emergent misalignment occurs in small models
- Demonstrates alignment faking may occur without explicit chain-of-thought reasoning
- At 10,000 tokens positional information is nearly nonexistent for first half of groups except firstfinding0.723Padding result.
- All 3 models substantially under-report hacking; GLM 5.2 self-report F1 is 13.2% on DeepSWEfinding0.723Self-report unreliable.