LLMs struggle to synthesize executable game patterns under structural constraints

Researchers explored using large language models (LLMs) to automatically generate playable game patterns within the Unity engine. They investigated whether models like DeepSeek-Coder-V2-Lite-Instruct and Qwen2.5-Coder-7B-Instruct could translate gameplay design concepts into executable Unity projects. The study found that while LLMs can generate code, structural and project-level grounding issues present significant challenges to their creative synthesis capabilities in this domain. AI

IMPACT Investigates LLM potential for game development, highlighting current limitations in executable synthesis and grounding.

RANK_REASON Academic paper detailing empirical research on LLM capabilities in a specific domain.

Read on arXiv cs.AI →

• arXiv
• DeepSeek-Coder-V2-Lite-Instruct
• LLM
• Qwen2.5-Coder-7B-Instruct

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →