Hierarchical Expert Prompt for Large-Language-Model: An Approach Defeat Elite AI in TextStarCraft II for the First Time (2502.11122v1)

Abstract: Since the emergence of the LLM, LLM has been widely used in fields such as writing, translating, and searching. However, there is still great potential for LLM-based methods in handling complex tasks such as decision-making in the StarCraft II environment. To address problems such as lack of relevant knowledge and poor control over subtasks of varying importance, we propose a Hierarchical Expert Prompt (HEP) for LLM. Our method improves the understanding of game situations through expert-level tactical knowledge, improving the processing quality of tasks of varying importance through a hierarchical framework. Our approach defeated the highest level (Elite) standard built-in agent in TextStarCraft II for the first time and consistently outperformed the baseline method in other difficulties. Our experiments suggest that the proposed method is a practical solution for tackling complex decision-making challenges. The replay video can be viewed on https://www.bilibili.com/video/BV1uz42187EF and https://youtu.be/dO3PshWLV5M, and our codes have been open-sourced on https://github.com/luchang1113/HEP-LLM-play-StarCraftII.

• The paper presents a novel Hierarchical Expert Prompt (HEP) technique that infuses expert tactical knowledge into LLMs for real-time strategic decision-making.
• It employs two key components—Expert Tactic Prompt (ETP) and Hierarchical Decision Prompt (HDP)—to efficiently manage resources and prioritize actions.
• Experimental results demonstrate a 75% win rate against Elite AI, highlighting HEP's potential to advance AI decision systems in complex environments.

A Formal Analysis of "Hierarchical Expert Prompt for Large-Language-Model: An Approach Defeat Elite AI in TextStarCraft II for the First Time"

The paper "Hierarchical Expert Prompt for Large-Language-Model: An Approach Defeat Elite AI in TextStarCraft II for the First Time" presents a novel methodology designed to enhance LLMs’ abilities in complex decision-making tasks, using StarCraft II as the experimental platform. The proposed approach, Hierarchical Expert Prompt (HEP), demonstrates how expert-level tactical knowledge can be systemically infused into LLMs to address challenges such as nuanced resource management and strategic decision-making in real-time strategy environments.

Technical Approach

The authors introduce a structured framework, HEP, encompassing two primary components: Expert Tactic Prompt (ETP) and Hierarchical Decision Prompt (HDP). The ETP equips the LLM with a repository of predefined tactics which defines key buildings, technologies, forces, timing, and scenarios for each strategy. This facilitates adaptive strategy selection based on real-time gameplay analysis, which aims to improve situational recognition and decision-making granularity.

HDP further stratifies potential actions within the game into prioritized and routine tasks. This hierarchical organization is pivotal in ensuring that decisions of higher importance, such as Nexus construction and resource management, are addressed first, effectively optimizing resource allocation and strategic planning under dynamic game conditions.

Experimental Results

The experimental setup pits the Hierarchical Expert Prompt method against varying difficulties of AI-controlled adversaries within the TextStarCraft II environment. Remarkably, HEP-LMM demonstrated superior performance, achieving the first-known victory against the Elite level AI. Quantitative results showed a substantial improvement in success rates over previous benchmark methods, particularly highlighting a 75% win rate against VeryHard AI—a previously unsurmounted challenge for LLM-based systems.

Notably, the computational cost of implementing HEP was examined and found to be in a reasonable range, with only a marginal increase in processing time and token usage, indicating efficiency despite the enhanced decision-making capabilities.

Implications and Future Directions

The implications of this research are significant for the domain of AI-driven decision systems. The Hierarchical Expert Prompt strategy could pave the way for more sophisticated AI architectures that integrate domain-specific expert knowledge tightly with real-time decision-making algorithms, potentially applicable to broader fields requiring complex, multi-layered decision frameworks.

Future developments may explore the scalability of HEP in other AI challenge realms beyond StarCraft II, such as complex simulation environments and real-world strategic applications. Another promising direction could involve refining tactic libraries with diverse strategic paradigms across different contexts, further enhancing adaptability and strategic depth. Additionally, exploring the integration of other machine learning paradigms, like reinforcement learning, with HEP could yield hybrid systems capable of even more nuanced and potent tactical expertise.

In conclusion, this paper provides a meaningful step forward in leveraging LLMs for complex decision-making tasks, combining structured tactical knowledge with hierarchical decision logic to achieve tasks traditionally requiring extensive training and resource usage. The accomplishments and methodical advancements presented here represent a valuable contribution to the intersection of AI, LLMs, and strategic game theory.