Self-correction on Llama-3.3-70B math n=30 (failure pool)
86.7Correction RateL_MEMORY (Source-Conditioned Role Relabeling)
Evaluation Results
| Method | Links | |
|---|---|---|
| L_MEMORY (Source-Conditioned Role Relabeling)Protocol=L_MEM.2026.06 | 86.7 | |
| ReflexionProtocol=Reflex.2026.06 | 6.7 | |
| auditProtocol=audit-only baseline2026.06 | 3.3 | |
| Self-RefineProtocol=S.-Refine2026.06 | 0 | |
| Chain-of-Verification (CoVe)Protocol=CoVe2026.06 | 0 |