Rethinking Fine-Tuning when Scaling Test-Time Compute: Limiting Confidence Improves Mathematical Reasoning Paper • 2502.07154 • Published Feb 11 • 1
