arXiv cs.CL· Lecheng Kong, Like Hui, Haitao Mao, Jun Huan·· 12 小时前AI 评分34
Consilience：面向无验证器测试时扩展的置信度轨迹选择框架
Consilience for Verifier-Free Test-Time Scaling
AI 导读
论文提出 consilience 框架，用于无验证器测试时扩展（VF-TTS），指出仅按置信度排序的现有方法在复杂任务上会灾难性失效——初始置信度普遍偏高往往意味着探索失败、偏向"自信的错误答案"。该框架通过组合指标惩罚高初始置信度、严格要求最终确定性，在研究生级数学题与自由形式代码生成实验中优于现有基线，论文已被 NeurIPS 2026 接收。
正文 · 原文
Abstract:Test-time scaling often uses an external verifier, such as compilers and test cases in coding or trained value functions in robotics applications, to obtain high-quality rollouts. Verifier-free test-time scaling (or VF-TTS) is gaining extensive attention as a mechanism to enhance Large Language Model (LLM) reasoning, primarily because we do not have access to such high-quality verifiers in many real-world applications. Among existing VF-TTS methods, confidence-based VF-TTS methods, which compute and rank rollouts solely by confidence, are particularly promising. Such methods introduce near-zero overhead for sample evaluation and require minimal access to internal model states, making the methods highly flexible across models and tasks.
In this paper, we demonstrate a critical limitation of existing confidence-based VF-TTS methods by showing that such methods catastrophically break down on complex tasks. We observe a very interesting phenomenon: uniformly high confidence frequently indicates a failure to explore, favoring confidently wrong answers. To address this, our core insight is that robust cognitive search requires a specific confidence trajectory pattern: such methods perform exploratory branching at the beginning, as manifested by low initial confidence, and converge to a high final confidence solution. To implement this insight, we introduce consilience, a novel selection framework that explicitly evaluates the temporal asymmetry of confidence in reasoning. We operationalize this via a combinatorial metric that actively penalizes high initial confidence while strictly demanding final certainty. Extensive experiments covering both graduate-level mathematics problems and free-form code generation demonstrate that consilience effectively outperforms existing baselines, validating our novel perspective on completion confidence.
| Subjects: | Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2608.09898 [cs.CL] |
| (or arXiv:2608.09898v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2608.09898 arXiv-issued DOI via DataCite |
|
| Journal reference: | Published at NeurIPS 2026 |
Submission history
From: Lecheng Kong [view email]
[v1]
Mon, 10 Aug 2026 17:45:44 UTC (291 KB)
[v2]
Fri, 30 Oct 2026 18:08:17 UTC (82 KB)
来源：arXiv cs.CL · arxiv.org