Abstract:Inference time optimization techniques, such as repeated sampling, have significantly advanced the reasoning capabilities of Large Language Models (LLMs). However, the critical role of model uncertainty remains largely underexplored in these optimization strategies. In this paper, we investigate the dynamics of confidence along reasoning trajectories and for first time reveal a surprising and unique pattern: correct answer traces tend to exhibit confidence improvement over time (positive confidence gain), while incorrect traces show attenuated or declining confidence as reasoning proceeds. Based on this observation, we propose Confidence Dynamic Gain (CDG) based voting, which incorporates how the confidence trajectory of the response evolves along the reasoning chain. Experiments across four open-source architectures (DeepSeek-R1, gpt-oss, Gemma-3, Qwen-QwQ) on the AIME24/25, HMMT25, and BRUMO25 benchmarks demonstrate that CDG yields a significant performance boost over baselines. These results demonstrate that our method provides a robust discriminative signal for improving answer selection in LLM reasoning. We also provide theoretical insights for this phenomenon. Code will be released at this https URL.
| Comments: | Published in ICML 2026 |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2605.25244 [cs.CL] |
| (or arXiv:2605.25244v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.25244 arXiv-issued DOI via DataCite (pending registration) |
From: Jiayun Wang [view email]
[v1]
Sun, 24 May 2026 20:04:19 UTC (296 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。