Logic Agent: Enhancing Validity with Logic Rule Invocation
Keywords: reasoning, large language models, logic, logic reasoning
TL;DR: We present the Logic Agent (LA) framework to boost the validity of reasoning in LLMs through strategic logic function calls.
Abstract: Chain-of-Thought (CoT) prompting has become a key strategy for enhancing the inferential abilities of large language models (LLMs) in reasoning tasks. However, it often struggles with ensuring reasoning validity and maintaining informativeness. This paper presents the Logic Agent (LA), a novel framework designed to boost the validity of reasoning in LLMs through strategic logic function calls. Distinct from traditional methods, LA converts LLMs into dynamic agents that apply propositional logic rules, transforming natural language inputs into structured logical forms. The agent utilizes a robust suite of predefined functions to guide the reasoning process effectively. This approach can enhance the structured and coherent generation of reasoning outputs, improving their interpretability and logical consistency. Through detailed experiments, we showcase LA's ability to adapt across different LLM sizes, significantly enhancing the accuracy of complex reasoning tasks across various domains.
Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8330
Loading