Best-of-200 with self-verification lifts Gemini 1.5 Pro past o1-Preview on several reasoning benchmarks and reveals that more samples make verification easier.
Never shorten your responses
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Sample, Scrutinize and Scale: Effective Inference-Time Search by Scaling Verification
Best-of-200 with self-verification lifts Gemini 1.5 Pro past o1-Preview on several reasoning benchmarks and reveals that more samples make verification easier.