Recently, the Chinese Academy of Sciences and Wang Jun’s team collaborated to launch TextStarCraftII, an eye-catching AI project designed to conquer the complex game of StarCraft II. This project uses the large model LLMAgent to demonstrate strategic capabilities in the game that surpass AlphaStar, including danger prediction, flexible switching of arms, and decision-making methods that are closer to humans. The team improved LLM's decision-making efficiency through the innovative Chain of Summarization method, and designed an exquisite prompt word system to enhance real-time decision-making and long-term strategic planning capabilities. This breakthrough research sets a new benchmark for the application of artificial intelligence in complex strategy games, and also provides valuable experience and reference for future artificial intelligence development.
The article focuses on:
Facing the challenge of StarCraft II, the Chinese Academy of Sciences and Wang Jun's team jointly released TextStarCraftII. This product uses the large model LLMAgent to demonstrate danger prediction, troop transformation and human-like strategies in StarCraft II that surpass AlphaStar. The new method ChainofSummarization is used to improve the decision-making ability of LLM, and a prompt word system is designed to improve real-time decision-making and long-term planning capabilities. Detailed information can be found in [paper](https://arxiv.org/pdf/2312.11865.pdf) and [project address](https://github.com/histmeisah/Large-Language-Models-play-StarCraftII).The success of TextStarCraftII marks the significant progress of artificial intelligence in the field of complex strategy games. Its technological innovation and strategic improvement provide new directions for future AI development. For more technical details, please visit the paper and project addresses provided for more information.