Reproduce “LLM-Based Authoring of Agent-Based Narratives through Scene Descriptions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Vinayak Regmi, Christos Mousas
Researchers at Purdue University developed a system that leverages lightweight Large Language Models to translate natural language scene descriptions into executable action sequences for virtual agents in a real-time 3D environment. The method successfully generated structurally valid and semantically coherent multi-agent narratives across varying complexities, with ChatGPT consistently outperforming other models in response time and consistency.
Key insights
- Even lightweight LLMs can reliably convert high-level natural language scene descriptions into executable, multi-agent behaviors within a real-time 3D environment.
- The approach fundamentally shifts narrative authoring from low-level scripting to intuitive scene composition, making it accessible to non-technical users.
- Meticulous prompt engineering and structured metadata are essential for LLMs to generate valid and coherent agent coordination plans, including object dependency management.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.