Reasoning models shift from implied to literal meanings when thinking longer

Entertain, then Cancel: How Thinking Makes Reasoning Models Read Some and Or Literally

Artificial IntelligenceComputation and Language

Summary

When people hear a statement like "some students passed," they usually assume not all of them did, but reasoning models can interpret this differently depending on how long they "think." The authors studied these models by changing only how long they process information, finding that more thinking makes models stick to the literal meanings of words like "some" and "or," canceling the assumed extra inferences people usually make. They show that models first consider the more specific implied meaning but often reject it after further thought. This discovery suggests evaluations should check how models interpret language, not just how accurate their answers are.

What this means in practice

Authors

Zhizhuo Kou, Jian Yang, Pengfei Hao, Junyu Luo, Jingshu Peng, Sirui Han, Yike Guo

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.
View PDFOpen arXiv