arXivBingnan Xiao, Chenhao Yang, Bingcong Li, Wei Ni, Xin WangTue, Oct 6, 2026, 8:43 AM PDT
score 16.7
Per-query sharpening boosts LLM reasoning without retraining
Original: Adaptive Power Sampling for LLM Reasoning
Source: arxiv.org ↗
Writing ELI5 summary…