DAHS and BHA reduce distribution sharpening in math RLVR, yielding gains in pass@1 and pass@2048 on AIME benchmarks by aligning hints to student responses and gradually removing them.
+” key is pressed, f× =operation performed when the “×
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Mitigating Distribution Sharpening in Math RLVR via Distribution-Aligned Hint Synthesis and Backward Hint Annealing
DAHS and BHA reduce distribution sharpening in math RLVR, yielding gains in pass@1 and pass@2048 on AIME benchmarks by aligning hints to student responses and gradually removing them.