Intervene When It Doubts: Conjunction-Guided Interactive Reasoning

Large Reasoning Models (LRMs) excel at complex reasoning but suffer from inefficient reasoning, notably overthinking and overshoot. These issues stem from excessive or misdirected reasoning triggered by the model's "doubt", which manifested as self-validation and exploratory extension, thereby increasing computational cost and degrading performance. Existing efficientreasoning methods regulate reasoning via internal signals or static schedules, but they are not tailored to the doubt-related characteristics of LRMs. To address this, we propose a Conjunction-Guided Intervention (CGI) reasoning framework that intervenes when the model shows signs of doubt. Our key insight is that overthinking and overshoot in LRMs arise from conjunctiontriggered extensions where LRMs signal "doubt" through transitional conjunctions, and then extend redundant self-validation or exploration without timely state-based correction. Building on this insight, CGI pauses reasoning at conjunction-based markers of doubt and injects external state-based feedback, adaptively extending or terminating reasoning to reduce redundancy while preserving accuracy. The feedback is generated through criterion-based evaluation of rationality and completeness and comes from either human or LLM proxies. We train the target model with Group Relative Policy Optimization (GRPO) to adapt to the interactive reasoning mode. Experiments show that our framework achieves a superior balance between accuracy and reasoning length.