| title | Entertain, then Cancel: How Thinking Makes Reasoning Models Read Some and Or Literally |
|---|---|
| authors | Zhizhuo Kou, Jian Yang, Pengfei Hao, Junyu Luo, Jingshu Peng, Sirui Han, Yike Guo |
| date | 2026-10-15 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2610.21204 |
| affiliation | HUAWEI Noah's Ark Lab |
| upvotes | 14 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| 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 some (not all) and or (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. |
cite
@misc{kou2026entertain,
title={Entertain, then Cancel: How Thinking Makes Reasoning Models Read Some and Or Literally},
author={Zhizhuo Kou and Jian Yang and Pengfei Hao and Junyu Luo and Jingshu Peng and Sirui Han and Yike Guo},
year={2026},
eprint={2610.21204},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2610.21204}
}