arXiv cs.AI· Zhizhuo Kou, Jian Yang, Pengfei Hao, Junyu Luo, Jingshu Peng, Sirui Han, Yike Guo·· 2 天前AI 评分40
思考如何让推理模型按字面理解"some"和"or"：Entertain, then Cancel
Entertain, then Cancel: How Thinking Makes Reasoning Models Read Some and Or Literally
AI 导读
研究发现，增加思考量会让推理模型把"some"（非全部）和"or"（非两者）的会话含义取消，转而按字面逻辑理解，这一现象在多数被测开源推理模型和 GPT-5 中均成立，而间接回答等依赖上下文的推理反而改善。推理轨迹显示模型先接受丰富含义、随后将其取消；一种在推理早期阶段作答的探索性方法可在不损失字面准确率的前提下恢复这些含义。
正文 · 原文
Abstract:When a speaker says ``some students passed'', listeners infer that not all of them did. Reasoning models now let users decide how long a model thinks before it answers, but we do not know what extra thinking does to such everyday inferences. We study this directly: we keep a model's weights fixed, change only how much it thinks, and record which reading it chooses. A second tool cuts the model's own reasoning at successive points and asks for an answer, showing where the reading changes. Thinking makes models read logical words as logic: it cancels the implicatures of \emph{some} (\emph{not all}) and \emph{or} (\emph{not both}) in most tested open reasoning models and in GPT-5, while context-driven inferences such as indirect answers improve. The trace shows why: models first entertain the enriched reading and cancel it later. Answering from that early reading, an exploratory inference-time remedy, recovers these implicatures without losing literal accuracy. Evaluations of reasoning models should measure how they interpret, not only how accurate they are.
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.21204 [cs.AI] |
| (or arXiv:2610.21204v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.21204 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jian Yang [view email]
[v1]
Thu, 15 Oct 2026 04:01:59 UTC (1,716 KB)
来源：arXiv cs.AI · arxiv.org