---
id: 20260618-T0-12
title: "CoRA：通过置信度-理由对齐提升大模型推理可靠性"
title_en: "CoRA: Improving LLM Reasoning Reliability via Confidence-Rationale Alignment"
url: https://ai.daily.yangsir.net/daily/20260618-T0-12
issue_date: 2026-06-18
publish_date: 2026-06-17T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2606.14961
---

# CoRA：通过置信度-理由对齐提升大模型推理可靠性

新研究CoRA解决了大模型链式推理中高置信度与不可靠理由的矛盾问题。研究发现LLM在伴随理由看似合理但支撑不足时仍会给出高置信度答案，导致误导。CoRA通过置信度-理由对齐机制，确保模型给出高置信度答案时，其推理过程也得到充分支撑，提升推理可靠性。

## English Version

**CoRA: Improving LLM Reasoning Reliability via Confidence-Rationale Alignment**

New research CoRA addresses the contradiction between high confidence and unreliable reasoning in LLM chain-of-thought. The study found that when LLMs' reasoning chains seem plausible but lack proper support, they still output high-confidence answers, leading to potential misinformation. CoRA introduces confidence-rationale alignment to ensure that when a model expresses high confidence, its reasoning process is well-supported, improving overall reasoning reliability.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2606.14961)

**详情页**：https://ai.daily.yangsir.net/daily/20260618-T0-12

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*