arXiv cs.AI· Huaman Sun, Dingcheng Wang, Jason Hartline, Jessica Hullman·· 11 小时前AI 评分31
诊断并改进大语言模型的概率推理能力
Diagnosing and Improving Probabilistic Reasoning in Large Language Models
AI 导读
研究提出一个决策论框架，将 LLM 的决策损失分解为"从证据形成准确信念"和"将信念转化为优化给定效用函数的行动"两部分，并用已知真值的合成基准刻画前沿与开源模型的概率推理表现。针对信念、决策或两者的 RL 干预在三个领域测试显示：单独针对某一成分会重新分配决策损失，改善目标成分但未必迁移到其他成分；联合训练两者可同时提升，但依赖训练与评估格式匹配。
正文 · 原文
Abstract:Large language models (LLMs) are increasingly proposed as decision assistants who must reason probabilistically from available evidence under explicit decision costs. We propose a decision-theoretic framework that decomposes LLMs' decision loss into two components: forming accurate beliefs from provided evidence and translating those beliefs into actions that optimize a provided utility function. Using a synthetic benchmark with known ground truth, we apply the decomposition to characterize probabilistic reasoning in frontier and open-sourced models. We further evaluate whether RL interventions targeting beliefs, decisions, or both improve these components across three domains, whether improvements transfer across components and elicitation formats, and whether decision performance can improve without improvement in belief formation. We find that targeting one component of probabilistic reasoning redistributes decision loss, improving the target without necessarily transferring to others, and that jointly targeting belief formation and decision-making improves both but hinges on matched formats between training and evaluation.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.38005 [cs.AI] |
| (or arXiv:2609.38005v3 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2609.38005 arXiv-issued DOI via DataCite |
Submission history
From: Huaman Sun [view email]
[v1]
Tue, 29 Sep 2026 16:55:50 UTC (508 KB)
[v2]
Tue, 6 Oct 2026 18:09:00 UTC (501 KB)
[v3]
Thu, 5 Nov 2026 02:21:56 UTC (515 KB)
来源：arXiv cs.AI · arxiv.org