| dc.description.abstract | This study explores the application of Large Language Models (LLM) in collaborative environments, emphasizing the integration of prompt engineering and the "Chain of Thought" method to mitigate hallucination in text generation, and highlighting the importance of cross-cultural collaboration. Introducing an AI agent that provides meeting summaries, guides thought, resolves conflicts, and evaluates its impact on collaboration efficiency, personalized interpretation, and conflict mediation.
Users can interact indirectly with this AI agent by creating a website with a user interface. Through this UI, users can intuitively evaluate the AI agent’s performance in collaboration. The presence of this UI provides users with a channel to comprehensively understand the AI agent’s functionality and effectively evaluate its collaborative performance. Experimental results demonstrate the AI agent’s success in streamlining meeting processes, fostering deeper discussions, and excelling at challenges such as reducing the number of conflicts over time and team collaboration. Specifically, the AI agent effectively minimizes errors, meaningless content, or situations divorced from reality, thereby improving the quality of the generated text. In addition, the AI agent’s guidance and conflict mediation capabilities contribute to smoother and more efficient meetings. However, areas for improvement in conflict resolution and bias reduction are identified, with the AI agent’s performance in these aspects rated as moderate, indicating potential avenues for further optimization.
In summary, the experiment results support the value of prompt engineering in collaborative scenarios and underscore the potential benefits of AI agents in enhancing collaboration efficiency and quality. Future research directions include further optimization of conflict resolution capabilities and reduction of potential biases. Recognizing the challenges of computational resources and runtime, future efforts should prioritize reducing computational costs, optimizing runtime, and expanding the system’s input methods to include speech input, thereby increasing user diversity and convenience. | |
