Deep Interaction: An Efficient Human-AI Interaction Method for Large Reasoning Models
2026-07-15 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors study how to fix mistakes that large language models make when reasoning through complicated problems step-by-step. Instead of asking the model to generate a whole new answer or pointing out errors without precise fixes, they propose a method called Deep Interaction, which lets users directly edit the parts of the answer that are wrong while keeping the right parts intact. This edited reasoning is then refined into a prompt that guides the model to follow the corrected steps. Their tests show this method is more successful and uses fewer words compared to existing ways of fixing mistakes in STEM problem-solving.
Chain-of-Thought reasoninglarge language modelshuman interventionDeep Interactionreasoning errorsSTEM tasksprompt engineeringerror correctiontoken efficiency
Authors
Hefeng Zhou, Jinxuan Zhang, Jiong Lou, Yuxin Liu, Chaochao Lu, Jingjing Qu, Jie Li
Abstract
The emergence of Chain-of-Thought (CoT) reasoning has significantly enhanced the ability of large language models (LLMs) to tackle complex, multi-step tasks. However, when errors occur, current interaction approaches typically involve re-generating another response that may make mistakes again, or users laboriously flag the faulty step in follow-up turns that may get responses <You are right, I made a mistake here> followed by similar errors recurring. To address this issue, we propose an efficient human intervention mechanism for precisely correcting reasoning errors in LLMs, termed Deep Interaction. Our approach enables direct editing of the original response, allowing erroneous parts to be corrected while preserving accurate reasoning steps. We refine the edited CoT into a distilled prompt, which then steers the LLM along the corrected reasoning path. Experimental results show that our method achieves over a 25% improvement in correction success rate and reduces token usage by approximately 40% on STEM tasks reasoning compared to baseline approaches.