InnoChat: A Heuristic Teaching Large Model for Innovation Thinking Training Based on Logical Chains and Prompt Tuning
Weisen Zhong, Feihong Ye, Chaocheng Zhong, Runcong Cai, Zehui Zhan · 2025
Cultivating innovative thinking is a critical topic in the field of AI-driven education. However, existing large language models (LLMs) still face significant limitations in innovation training, including a narrow scope of knowledge transmission and insufficient depth of interaction. To address these challenges, this study develops InnoChat, an innovative thinking training system based on GPT-4, introducing a dual-track enhancement mechanism. By integrating logical chain fine-tuning and prompt optimization, InnoChat significantly improves its capability to guide innovation thinking. A randomized controlled experiment was conducted, involving 30 university students in comparative tests. The results demonstrate that InnoChat achieved a Creativity Stimulation Score of 9.05 (SD = 0.62), significantly outperforming GPT-4, which scored 8.20 (SD = 0.80). Additionally, InnoChat exhibited notable advantages in thinking depth and practical value (p < 0.001). Furthermore, 93.5% of participants reported that InnoChat effectively stimulated their innovative thinking, while 89.2% found the interaction process natural and engaging. These findings highlight a paradigm shift, transforming AI from a mere knowledge provider into an active facilitator of cognitive training. This study not only validates the feasibility of AI-driven innovation thinking training but also offers theoretical and practical insights for developing next-generation intelligent innovation education tools.