让大模型自己辩论来判断答案可信度，只输出有把握的结果。
Trust but Verify: Prover-Verifier Deliberation for Selective LLM Prediction
- 用两个模型轮流当论证方和验证方，通过问答对话评估答案可靠性。
- 在GPQA钻石题集上，高置信答案正确率比普通答案高出约30个百分点。
- 适合需要精准决策的场景，如医疗、法律等高风险应用。
可靠地判断语言模型何时正确，与模型本身正确同样重要。我们提出证明者-验证者思辨（PVD），一种基于交互式证明理论的推理时协议，用于选择性预测：该协议生成答案和结构化置信度判断，使系统可报告高置信答案并放弃不确定情况。在每次对话中，证明者通过可检验的子命题辩护候选答案，验证者则发出针对性挑战，并返回【接受】、【挑战】或【拒绝】。由于冻结的语言模型作为不完美的证明者和验证者，在噪声通道上运行，形式上的正确性和完备性无法保证；因此，我们通过覆盖率-精确率行为对协议进行经验刻画。主实验使用Claude Sonnet 4.6作为证明者，Claude Haiku 4.5作为验证者，在GPQA Diamond上测试。在未修改答案的情况下被接受的题目（称为接受+无更改，ANC），构成高置信子集；我们通过其精确率和覆盖率评估该子集。ANC有效区分了可靠与不可靠答案，相较非ANC补集产生约30个百分点的高置信精确率差距。跨GPT与Gemini组合的鲁棒性实验表明，高置信精确率可在模型家族间转移，而验证者严格程度和领域能力主要决定选择差距大小。在‘人类最后的考试’数据集上，较弱的证明者-验证者配对可能导致ANC信号崩溃甚至反转，揭示了验证者超出有效区域时的实际失效模式。与自洽性、通用自洽性、多代理辩论及Reflexion的对比显示，证明者-验证者思辨为选择性预测提供了独特的论点可辩护信号。
原文摘要 · Abstract (English)
Reliably knowing when a language model is correct is almost as important as being correct. We introduce prover-verifier deliberation (PVD), an inference-time protocol grounded in interactive proof theory, as a mechanism for selective prediction: the protocol produces both an answer and a structured confidence verdict, allowing a system to report high-confidence answers while abstaining on uncertain cases. In each dialogue, a prover defends a candidate answer through checkable sub-claims while a verifier issues targeted challenges and returns \textsc{Accept}, \textsc{Challenge}, or \textsc{Reject}. Because frozen language models are imperfect provers and verifiers operating over a noisy channel, formal soundness and completeness guarantees do not transfer; instead, we characterize the protocol empirically through its coverage-precision behavior. Our main experiment uses Claude Sonnet 4.6 as prover and Claude Haiku 4.5 as verifier on GPQA Diamond. Questions accepted with no answer revision, which we call Accept + No Change (ANC), are reported as the high-confidence subset; we evaluate this subset by its precision and coverage. ANC separates reliable from unreliable answers, yielding a $\sim$30pp HC-Prec gap over the non-ANC complement. Robustness experiments with GPT and Gemini pairings show that high HC-Prec can transfer across model families, while verifier strictness and domain competence largely determine the size of the selection gap. On Humanity's Last Exam, weaker prover-verifier pairings can collapse or invert the ANC signal, illustrating a practical failure mode when the verifier operates outside its effective region. Comparisons with self-consistency, universal self-consistency, multi-agent debate, and Reflexion suggest that prover-verifier deliberation supplies a distinct argument-defensibility signal for selective prediction.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。