finding
active
finding:policy-averages-63-4-0-4-entities-per-sequence-versus-base-63-6-0-3Policy averages 63.4±0.4 entities per sequence versus base 63.6±0.3
Not merely making fewer claims.
Source paper
extracted_from(2026) · Aaditya Vikram Prasad · Connor Watts · Jack Merullo · Dhruvil Gala +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Rules out overcautiousness.
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.
- Average KL between base and policy is 10–20% larger on Not Supported than Supported tokensfinding0.766Targeted updates.
- More cautious claims.
- Intervention action distribution.
- Basic SAE performance metrics.
- Spectrum of brittleness.
- Component of decomposition.
- Intervals miscalibrated.
- RLFR policy matches base on standard benchmarks (e.g., MMLU 67.2 vs 69.2, GSM8K 73.0 vs 72.3)finding0.711No capability degradation.