论文探讨强化学习如何改变大模型推理：可迁移性、覆盖度与缩放规律
摘要
alphaXiv 推介的一篇论文提出，强化学习主要改变大模型偏好哪些推理策略，而非教会全新的推理技能：强化某些策略可提升部分任务准确率，却可能压制对其他任务有用的策略并降低推理多样性。
摘要由模型根据原文撰写，以原文为准。
正文 · 原文
"How RL Reshapes LLM Reasoning: Transferability, Coverage, and Scaling Laws"
Does RL actually teach LLMs new reasoning skills or just make them better at using the ones they already have?
This paper suggests RL primarily changes which reasoning strategies models prefer. So by reinforcing certain approaches, it can improve accuracy on some tasks while suppressing useful strategies for others, and potentially reducing reasoning diversity.
On Qwen3-14B, RL reduces reasoning strategy diversity by 18%, but improves BoxNet Pass@1024 from 34% to 51%, showing that fewer strategies can still lead to broader problem-solving coverage.
来源：alphaXiv · x.com
引用原文
askalphaxiv. "How RL Reshapes LLM Reasoning: Transferability, Coverage, and Scaling Laws"[EB/OL]. (2026-10-27)[访问日期]. https://x.com/askalphaxiv/status/2114907436725232060.