finding
active
finding:rlfr-without-bon-achieves-56-6-1-9-overall-reductionRLFR without BoN achieves 56.6%±1.9% overall reduction
Ablation of BoN.
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
- Central thesis.
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.
- Ablation removing inline interventions.
- SFT on correction+retraction data gives 39.4% inline overall reduction and 9.6% not-inline, well below RLFRfinding0.780SFT baseline comparison.
- Topline result.
- Comparable to RLFR-NI.
- Establishes low-bar baseline showing personality control without intervention is poor
- RLFR policy matches base on standard benchmarks (e.g., MMLU 67.2 vs 69.2, GSM8K 73.0 vs 72.3)finding0.742No capability degradation.
- Even stylistic distribution.
- Inhibiting RM bias features reduces 9-1-1 mentions to 6% and voting to 0%; baseline 42% 9-1-1, 56% voting.finding0.733Law-schools example.