arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2512.02304cs.CL
收藏

验证何时有效?对LLM作为解决方案验证器的深入探讨

When Does Verification Pay Off? A Closer Look at LLMs as Solution Verifiers

Agentic Learning AI Lab, New York University(代理学习人工智能实验室,纽约大学)

Jack Lu, Ryan Teehan, Jinran Jin, Mengye Ren

AI总结:

研究探讨了在何种条件下验证有效,通过分析37个模型在9个基准测试中的表现,发现跨模型家族验证效果优于自验证,且验证收益随模型相似性增加而降低。

摘要:

大型语言模型(LLM)可以同时作为问题解决者和解决方案验证器,后者从解决者生成的候选答案中选择高质量答案。这引发了验证在解决者-验证者系统中何时有效的疑问。先前研究仅限于影响验证性能的因素,主要集中在自我验证,并未探讨解决者和验证者模型家族之间的关系或训练后推理的影响。为此,我们系统地研究了37个模型,涵盖多个家族、大小和基础与训练后变体,在9个基准测试中进行评估,涵盖逻辑推理、结构化谜题、符号计算、数学、常识、事实回忆和领域知识。为了支持我们的分析,我们引入并验证了验证收益指标,该指标预测了测试时基于验证器的拒绝采样带来的性能提升。我们的实验发现,1)跨模型家族的验证比自我验证或同一家族内的验证更有效,且一般来说,验证的益处随着解决者和验证者变得越来越相似而减少;2)训练后推理削弱了自我改进能力,但增强了跨家族改进;3)某些任务本质上更易于通过验证改进,尤其是数学和逻辑任务。

英文摘要:

Large language models (LLMs) can act as both problem solvers and solution verifiers, where the latter select high-quality answers from a pool of solver-generated candidates. This raises the question of under what conditions verification pays off in solver-verifier systems. Prior work has conducted only limited studies of the factors influencing verification performance, focusing primarily on self-verification and examining neither the relationship between solver and verifier model families nor the effects of reasoning post-training. To rectify this, we present a systematic study across 37 models spanning multiple families, sizes, and base vs. post-trained variants, evaluated on 9 benchmarks covering logical reasoning, structured puzzles, symbolic computation, mathematics, commonsense, factual recall, and domain knowledge. In order to support our analysis, we introduce and empirically validate verifier gain, a metric that predicts the performance improvements from test-time verifier-based rejection sampling. Our experiments find that 1) verification across model families is more effective than either self-verification or verification within the same family, and more generally that the benefits of verification decrease as the solver and verifier become more similar, 2) reasoning post-training weakens self-improvement abilities but strengthens cross-family improvement, and 3) some tasks are inherently more amenable to improvement through verification, particularly mathematical and logical tasks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑