AI RESEARCH
Reforming the Mechanism: Editing Reasoning Patterns in LLMs with Circuit Reshaping
arXiv CS.CL
•
ArXi:2603.06923v1 Announce Type: new Large language models (LLMs) often exhibit flawed reasoning ability that undermines reliability. Existing approaches to improving reasoning typically treat it as a general and monolithic skill, applying broad