#logical reasoning

This research contributes to the theoretical foundations of SMT solving, which is a core component of many automated reasoning and verification tools.

#logical reasoning #automated theorem proving

arXiv cs.AI@Yasmine Briefs, Christoph WeidenbachSource

Sep 3

PapersSelected73°09:07

Logical Validity Representations in Language Models

Models can represent logical validity internally without reliably expressing it in outputs, affecting how we evaluate AI reasoning capabilities.

#transformer #logical reasoning #language models #hidden states

arXiv cs.LG@Smitha Muthya Sudheendra, Jaideep SrivastavaSource

Aug 6

PapersSelected10:46

Chinese content同式异义：语言模型能否遵循模态逻辑规范？

这篇论文测了语言模型会不会按给定的模态逻辑规则推理，DeepSeek V4 Flash开推理模式后从4.4%直接飙到88.1%，反差很大，可以看看。

#DeepSeek #V4 Flash #logical reasoning

arXiv: DeepSeek@Réemi Andrieu, Damien SileoSource

Jul 31

Papers12:05

Chinese content选择性可信度限制信念更新

这篇 arXiv 论文给信念更新加了选择性机制，能处理部分输入不可信的情况，比旧方法更灵活。

#Katsuno-Mendelzon #logical reasoning

arXiv cs.AI@Theofanis Aravanis, Costas D. KoutrasSource

Jun 29

Papers10:12

Chinese contentCOCOLogic-V2：利用真实难负例识别逻辑不一致性

COCOLogic-V2 这个新数据集专测视觉推理，正反例分类特别细，模型在近边界上直接翻车，做可解释 AI 的可以看看。

#COCOLogic-V2 #reasoning models #vision understanding #logical reasoning

arXiv cs.LG@David Steinmann et al. (4)Source