[论文] When to Trust the Cheap Check: Weak and Strong Verification for Reasoning

论文概要

研究领域: LLM 作者: Shayan Kiyani, Sima Noorani, George Pappas, Hamed Hassani 发布时间: 2026-02-19 arXiv: 2602.17633

中文摘要

LLM推理正越来越多地在更广泛的验证循环中展开。内部层面，系统使用廉价检查如自一致性或代理奖励（称为弱验证）；外部层面，用户检查输出并通过反馈引导模型直到结果可信（称为强验证）。本文通过弱-强验证策略形式化这一张力：决定何时基于弱验证接受/拒绝，何时转交强验证。研究表明最优策略具有双阈值结构，校准度和锐度决定了弱验证器的价值。在此基础上，作者开发了一种在线算法，可在不对查询流、语言模型或弱验证器做任何假设的情况下，可证明地控制接受和拒绝错误。

原文摘要

Reasoning with LLMs increasingly unfolds inside a broader verification loop. Internally, systems use cheap checks, such as self-consistency or proxy rewards, which we call weak verification. Externally, users inspect outputs and steer the model through feedback until results are trustworthy, which we call strong verification. We formalize this tension through weak-strong verification policies, which decide when to accept or reject based on weak verification and when to defer to strong verification. We show that optimal policies admit a two-threshold structure and that calibration and sharpness govern the value of weak verifiers. Building on this, we develop an online algorithm that provably controls acceptance and rejection errors.

--- *自动采集于 2026-06-24*

#论文 #arXiv #LLM #小凯